Fourier Analysis on Number Fields (Graduate Texts in Mathematics) (v. 186)

Graduate Texts in Mathematics Dinakar Ramakrishnar Robert J.Valenza Fourier Analysis on Number Fields Springer Gradu...

Author: Dinakar Ramakrishnan | Robert J. Valenza

74 downloads 914 Views 7MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Graduate Texts in Mathematics Dinakar Ramakrishnar Robert J.Valenza

Fourier Analysis on Number Fields

Springer

Graduate Texts in Mathematics

186

Editorial Board S. Axler F.W. Gehring K.A. Ribet

Springer New York

Berlin Heidelberg Barcelona Hong Kong London Milan Paris Singapore

Tokyo

Graduate Texts in Mathematics I

TAKEUTVZARLNG. Introduction to

2

Axiomatic Set Theory. 2nd ed. OXTOBY. Measure and Category. 2nd ed.

33 34

HtRSCH. Differential Topology. SPrrzER. Principles of Random Walk. 2nd ed.

SCHAEFER. Topological Vector Spaces. HILTON/STAMMBACH. A Course in

35

ALEXANDER/WERMER. Several Complex

4

36

Variables and Banach Algebras. 3rd ed. KELLEY/NAMIOKA et al. Linear

5

Homological Algebra. 2nd ed. MAC LANE. Categories for the Working Mathematician. 2nd ed.

37

3

6

HUGHEs/PIPER. Projective Planes.

7

SERRE. A Course in Arithmetic. TAKEUn/ZARING. Axiomatic Set Theory. HUMPHREYS. Introduction to Lie Algebras

8 9

Topological Spaces. MONK. Mathematical Logic. 38 GRAUERT/FRm_SCHE. Several Complex Variables. 39 ARVESON. An Invitation to C'-Algebras. 40 KEMENY/SNELUKNAPP. Denumerable

Markov Chains. 2nd cd.

and Representation Theory. 10

COHEN. A Course in Simple Homotopy Theory.

II

CONWAY. Functions of One Complex Variable 1. 2nd ed.

12

BEALS. Advanced Mathematical Analysis.

13

ANDERSON/FULLER. Rings and Categories

41

APOSTL. Modular Functions and Dirichlet Series in Number Theory. 2nd ed.

42

SERRE. Linear Representations of Finite

43

Groups. GILLMAN/JERISON. Rings of Continuous

of Modules. 2nd ed.

Functions. KENDIG. Elementary Algebraic Geometry. LoEVE. Probability Theory 1. 4th ed.

and Their Singularities.

44 45

15

BERBERtAN. Lectures in Functional

46 LoEvE. Probability Theory 11. 4th ed. 47

16

Analysis and Operator Theory. WINTER. The Structure of Fields.

17

ROSENBLATT. Random Processes. 2nd ed.

18

HALMOS. Measure Theory. HALMOS. A Hilbert Space Problem Book.

48 SACHS/WU. General Relativity for Mathematicians.

14

19

GOLUBITSKY/GUILLEMIN. Stable Mappings

2nd ed. 20 HISEMOLLER. Fibre Bundles. 3rd ed. 21

HUMPHREYS. Linear Algebraic Groups.

22 BARNES/MACK. An Algebraic Introduction to Mathematical Logic. 23 GREUB. Linear Algebra. 4th ed. 24

HOLMES. Geometric Functional Analysis

and Its Applications. 25

HEWITT/STROMBERG. Real and Abstract

Analysis.

26

MANES. Algebraic Theories.

KELLEY. General Topology. 28 ZARISKU$AMUEL. Commutative Algebra. Vol.1. 27

Dimensions 2 and 3.

49 GRUENBERG/WEIR. Linear Geometry. 2nd ed. 50 EDWARDS. Fermat's Last Theorem. 51 KLtNGENBERG. A Course in Differential Geometry. 52 HARTSHORNE. Algebraic Geometry. 53 MANIN. A Course in Mathematical Logic. 54 GRAVER/WATKINS. Combinatorics with Emphasis on the Theory of Graphs. 55 BROWN/PEARCY. Introduction to Operator Theory I: Elements of Functional Analysis. 56 MASSEY. Algebraic Topology: An Introduction. 57

CROWEL/Fox. Introduction to Knot

58

Theory. KOBLITZ. p-adic Numbers, p-adic

29 ZAwsK1/SAMUE.. Commutative Algebra. Vol.11.

30 JACOBSON. Lectures in Abstract Algebra 1. Basic Concepts. 31 JACOBSON. Lectures in Abstract Algebra II. Linear Algebra. 32

JACOBSON. Lectures in Abstract Algebra

Ill. Theory of Fields and Galois Theory.

MOISE. Geometric Topology in

Analysis, and Zeta-Functions. 2nd ed. 59 60

LANG. Cyclolomic Fields. ARNOLD. Mathematical Methods in

Classical Mechanics. 2nd ed. 61

WHITEHEAD. Elements of Homotopy Theory.

(continued after index)

Dinakar Ramakrishnan Robert J. Valenza

Fourier Analysis on Number Fields

Springer

Dinakar Ramakrishnan Mathematics Department California Institute of Technology Pasadena, CA 91125-0001 USA

Editorial Board S. Axler Mathematics Department San Francisco State

University

San Francisco, CA 94132 USA.

Robert J. Valenza Department of Mathematics Claremont McKenna College Claremont. CA 91711-5903 USA

F.W. Gehring Mathematics Department East Hall University of Michigan Ann Arbor, MI 48109 USA.

K.A. Ribet

Mathematics Department University of California at Berkeley

Berkeley, CA 94720-3840

USA

Mathematics Subject Classification (1991): 42-01. I 1 F30 Library of Congress Cataloging-in-Publication Data Ramakrishnan, Dinakar. Fourier analysis on number fields / Dinakar Ramakrishnan, Robert J. Valenza. cm. - (Graduate texts in mathematics ; 186) p. Includes bibliographical references and index. ISBN 0-387-98436-4 (hardcover : alk. paper) 1. Fourier analysis. 2. Topological groups. 3. Number theory. 1. Valenza, Robert J., 1951. 11. Title. Ill. Series QA403.5.R327 1998 515 ".2433-dc21

98-16715

Printed on acid-free paper.

© 1999 Springer-Verlag New York, Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer-Verlag New York, Inc., 175 Fifth Avenue, New

York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use of general descriptive names, trade names, trademarks. etc., in this publication, even if

the former are not especially identified, is not to he taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone. Production managed by Terry Kornak: manufacturing supervised by Thomas King. Photocomposed copy provided by the authors. Printed and bound by Edwards Brothers. Ann Arbor. MI. Printed in the United States of America. 9 8 7 6 5 4 3 2

1

ISBN 0-387-98436-4 Springer-Verlag New York Berlin Heidelberg

SPIN 10659801

To Pat and Anand

To Brenda

Preface

This book grew out of notes from several courses that the first author has taught over the past nine years at the California Institute of Technology, and earlier at the Johns Hopkins University, Cornell University, the University of Chicago, and the University of Crete. Our general aim is to provide a modern approach to number theory through a blending of complementary algebraic and analytic perspectives, emphasizing harmonic analysis on topological groups. Our more

particular goal is to cover John Tate's visionary thesis, giving virtually all of the necessary analytic details and topological preliminaries-technical prerequisites that are often foreign to the typical, more algebraically inclined number theorist. Most of the existing treatments of Tate's thesis, including Tate's own, range from terse to cryptic; our intent is to be more leisurely, more comprehensive, and more comprehensible. To this end we have assembled material that has admittedly been treated elsewhere, but not in a single volume with so much detail and not with our particular focus.

We address our text to students who have taken a year of graduate-level courses in algebra, analysis, and topology. While our choice of objects and methods is naturally guided by the specific mathematical goals of the text, our approach is by no means narrow. In fact, the subject matter at hand is germane not only to budding number theorists, but also to students of harmonic analysis or the representation theory of Lie groups. We hope, moreover, that our work will be a good reference for working mathematicians interested in any of these fields. A brief sketch of each of the chapters follows.

(1) Topot.octc %1 GROUPS. The general discussion begins with basic notions

and culminates with the proof of the existence and uniqueness of Haar (invariant) measures on locally compact groups. We next give a substantial introduction to profinite groups, which includes their characterization as compact. totally disconnected topological groups. The chapter concludes with the elementary theory of pro-p-groups, important examples of which surface later in connection with local fields. (2) SOME REPRESENTATION THEORY. In this chapter we introduce the funda-

mentals of representation theory for locally compact groups, with the ultimate

viii

Preface

aim of proving certain key properties of unitary representations on Hilbert spaces. To reach this goal, we need some weighty analytic prerequisites, including an introduction to Gelfand theory for Banach algebras and the two spectral theorems. The first we prove completely; the second we only state, but with enough background to be thoroughly understandable. The material on Gclfand theory fortuitously appears again in the following chapter, in a somewhat different context. (3) DUALITY FOR LOCALLY COMPACT ABELIAN GROUPS. The main points here

are the abstract definition of the Fourier transform, the Fourier inversion formula, and the Pontryagin duality theorem. These require many preliminaries, including the analysis of functions of positive type, their relationship to unitary representations, and Bochner's theorem. A significant theme in all of this is the interplay between two alternative descriptions of the "natural" topology on the dual group of a locally compact abelian group. The more tractable description, as the compact-open topology, is presented in the first section; the other, which arises in connection with the Fourier transform, is introduced later as part of the proof of the Fourier inversion formula. We have been greatly influenced here by the seminal paper on abstract harmonic analysis by H. Cartan and R. Godement (1947), although we give many more details than they, some of which are not obvious-even to experts. As a subsidiary goal of the book, we certainly hope that our exposition will encourage further circulation of their beautiful and powerful ideas. (4) THE STRUCTURE OF ARITHMETIC FIELDS. In the first two sections the basics

of local fields, such as the p-adic rationals Q. are developed from a completely topological perspective; in this the influence of Weil's Basic Number Theory (1974) is apparent. We also provide some connections with the algebraic construction of these objects via discrete valuation rings. The remainder of the chapter deals with global fields, which encompass the finite extensions of Q and function fields in one variable over a finite field. We discuss places and completions, the notions of ramification index and residual degree, and some key points on local and global bases. (5) ADELES, IDELES, AND THE CLASS GROUPS. This chapter establishes the fun-

damental topological properties of adele and idcle groups and certain of their quotients. The first two sections lay the basic groundwork of definitions and elementary results. In the third, we prove the crucial theorem that a global field embeds as a cocompact subgroup of its adele group. We conclude, in the final section, with the introduction of the adele class group, a vast generalization of the ideal class group, and explain the relationship of the former to the more traditional ray class group. (6) A QUICK TOUR OF CLASS FIELD THEORY. The material in this chapter is not

logically prerequisite to the development of Tate's thesis, but it is used in our

Preface

ix

subsequent applications. We begin with the Frobenius elements (conjugacy classes) associated with unramified primes P of a global field F, first in finite Galois extensions, next in the maximal extension unramified at P. In the next three sections we state the Tchebotarev density theorem, define the transfer map for groups, and state, without proof, the Artin reciprocity law for abelian extensions of global and local fields, in terms of the more modern language of idele classes. In the fifth and final section, we explicitly describe the cyclotomic

extensions of Q and Q. and then apply the reciprocity law to prove the Kronecker-Weber theorem for these two fields. (7) TATE's THESIS AND APPLICATIONS. Making use of the characters and duality

of locally compact abelian groups arising from consideration of local and global

fields, we carefully analyze the local and global zeta functions of Tate. This brings us to the main issue: the demonstration of the functional equation and analytic continuation of the L-functions of characters of the idele class group. There follows a proof of the regulator formula for number fields, which yields the residues of the zeta function of a number field F in terms of its class number hF and the covolume of a lattice of the group OF of units, in a suitable Euclidean space. From this we derive the class number formula and, in consequence, Dirichlet's theorem for quadratic number fields. Further investigation of these L-functions-in fact, some rather classical analysis-next yields another fundamental property: their nonvanishing on the line Re(s)=1. Finally, as a most remarkable application of this material, we prove the following theorem of Hecke: Suppose that X and x' are idele class characters of a global field K and that XP XP' for a set of primes of positive density. Then X=FX' for some character y of finite order. One of the more parenthetical highlights of this chapter (see Section 7.2) is the explanation of the analogy between the Poisson summation formula for number fields and the Riemann-Roch theorem for curves over finite fields. We have given a number of exercises at the end of each chapter, together with hints, wherever we felt such were advisable. The difficult problems are often broken up into several smaller parts that are correspondingly more accessible. We hope that these will promote gradual progress and that the reader will take great satisfaction in ultimately deriving a striking result. We urge doing as many problems as possible; without this effort a deep understanding of the subject cannot be cultivated.

Perhaps of particular note is the substantial array of nonstandard exercises found at the end of Chapter 7. These span almost twenty pages, and over half of them provide nontrivial complements to, and applications of, the material developed in the chapter. The material covered in this book leads directly into the following research areas.

x

Preface

+ L -functions of Galois Representations. Following Artin, given a finitedimensional, continuous complex representation aof Gal(Q/Q), one associates an L-function denoted L(as). Using Tate's thesis in combination with a theorem of Brauer and abelian class field theory, one can show that this function has a meromorphic continuation and functional equation. One of the major open problems of modern number theory is to obtain analogous results for I-adic Galois representations ci, where I is prime. This is known to be true for Q arising from abelian varieties of CM type, and L(oi,s) is in this case a product of L-functions of idele class characters, as in Tate's thesis.

4 Jacquet-Langlands Theory. For any reductive algebraic group G [for instance, GL (F) for a number field F], an important generalization of the set of idele class characters is given by the irreducible automorphic representations ,r of the locally compact group G(AF). The associated L -functions L(n;s) are well understood in a number of cases, for example for GL,,, and by

an important conjecture of Langlands, the functions L(oi,s) mentioned above arc all expected to be expressible in terms of suitable L(n;s). This is often described as nonabelian class field theory.

4 The p-adic L -functions. In this volume we consider only complex-valued (smooth) functions on local and global groups. But if one fixes a prime p and replaces the target field C by C,,, the completion of an algebraic closure of Qp, strikingly different phenomena result. Suitable p-adic measures lead to p-adic-valued L-functions, which seem to have many properties analogous to the classical complex-valued ones.

4 Adelic Strings. Perhaps the most surprising application of Tate's thesis is to the study of string amplitudes in theoretical physics. This intriguing connection is not yet fully understood.

Acknowledgments Finally, we wish to acknowledge the intellectual debt that this work owes to H. Cartan and R. Godement, J.-P. Serre (1968, 1989, and 1997), A. Well, and,

of course, to John Tate (1950). We also note the influence of other authors whose works were of particular value to the development of the analytic background in our first three chapters; most prominent among these are G. Folland (1984) and G. Pedersen (1989). (See References below for complete bibliographic data and other relevant sources.)

Contents

PREFACE .....................................................................................................vii INDEX OF NOTATION ................................................................................ xv

1

TOPOLOGICAL GROUPS

1.1 Basic Notions ..................................................................................... 1 1.2 Haar Measure ..................................................................................... 9 1.3 Profinite Groups ............................................................................... 19 1.4 Pro -p-Groups ................................................................................... 36 Exercises ................................................................................................. 42

2

SOME REPRESENTATION THEORY

2.1 Representations of Locally Compact Groups ..................................... 46 2.2 Banach Algebras and the Gclfand Transform ................................... 50 2.3 The Spectral Theorems .................................................................... 60 2.4 Unitary Representations ................................................................... 73 Exercises ................................................................................................. 78 3

DUALITY FOR LOCALLY COMPACT ABELIAN GROUPS

3.1 The Pontryagin Dual ........................................................................ 86 3.2 Functions of Positive Type ............................................................... 91 3.3 The Fourier Inversion Formula ....................................................... 102 3.4 Pontryagin Duality ......................................................................... 118 Exercises ............................................................................................... 125

xii

4

Contents

THE STRUCTURE OF ARITHMETIC FIELDS

4.1 The Module of an Automorphism ................................................... 132 4.2 The Classification of Locally Compact Fields ................................. 140 4.3 Extensions of Local Fields .............................................................. 150 4.4 Places and Completions of Global Fields ........................................ 154 4.5 Ramification and Bases ..................................................................165 Exercises ...............................................................................................174

5

ADELES, IDELES, AND THE CLASS GROUPS

5.1 Restricted Direct Products, Characters, and Measures ..................... 180 5.2 Adeles, Ideles, and the Approximation Theorem ............................. 189 5.3 The Geometry of AK/K ...................................................................191 5.4 The Class Groups ...........................................................................196

Exercises ...............................................................................................208

6 A QUICK TOUR OF CLASS FIELD THEORY 6.1 Frobenius Elements ........................................................................214 6.2 The Tchebotarev Density Theorem .................................................219 6.3 The Transfer Map ...........................................................................220 6.4 Artin's Reciprocity Law ..................................................................222 6.5 Abelian Extensions of Q and Qp .....................................................226 Exercises ............................................................................................... 238

7

TATE'S THESIS AND APPLICATIONS

7.1 Local -Functions ...........................................................................243 7.2 The Riemann-Roch Theorem ..........................................................259 7.3 The Global Functional Equation .....................................................269 7.4 Hecke L-Functions ..........................................................................276 7.5 The Volume of CK and the Regulator ............................................281 7.6 Dirichlet's Class Number Formula ..................................................286 7.7 Nonvanishing on the Line Re(s)=1 .................................................289 7.8 Comparison of Hecke L-Functions ..................................................295 Exercises ...............................................................................................297

Contents

xiii

APPENDICES

Appendix A: Normed Linear Spaces A.1 Finite-Dimensional Notmed Linear Spaces .................................... 315 A.2 The Weak Topology ...................................................................... 317 A.3 The Weak-Star Topology............................................................... 319 A.4 A Review of LP-Spaces and Duality ............................................... 323

Appendix B: Dedekind Domains B.1 Basic Properties ............................................................................. 326 B.2 Extensions of Dedekind Domains .................................................. 334 REFERENCES ............................................................................................. 339

INDEX ........................................................................................................ 345

Index of Notation

Notation

Section

Interpretation

N,Z,Q

natural numbers, integers, and rational numbers, respectively

R,C

real and complex numbers, respectively nonnegative reals, positive reals identity map on the set S complement of the set S

R"

is Sc

Card(S) U S. supp(f)

T(X)

cardinality of the set S disjoint union of sets S.

support of a function f continuous (complex-valued) functions on a topological space X continuous functions with compact support positive elements of WA(X) with positive

Z/nZ q(n)

sup norm nonzero elements of a ring or field group of units of a ring A degree of a finite field extension K/F norm map for a finite field extension K/F; see also Section 6.4 trace map for a finite field extension K/F compositum of fields K and L integers modulo it Euler phi function

S1

the circle group

Wl

orthogonal complement of a subspace W

prw

orthogonal projection onto a subspace W

A*, K* AX

(K:F] NwF(x) trx7F(x)

KL

xvi Index of Notation

k[[t]]

k((t)) GL,,(k)

--

ring of formal power series in t with coefficients in the field k fraction field of k[it]]

group of invertible nxn matrices over k

SL (k) B'(X)

A.1

X"`

A.1

11(C')

Al

C with 1i norm

L(X)

A.4

measurable functions on X modulo agreement almost everywhere

Lo(X)

A.4

LP-space associated with a locally compact

- lip

AA

LP-norm

AS

B1

JK

B.2

PK

B.2

C/K

A(x,,...,x,)

B2 B2 B2

localization of a ring A at subset S set of fractional ideals of a global field K set of principal fractional ideals of K traditional class group of a global field K absolute norm map

A(B/A)

B.2

U, Rh, f

L1

nxn matrices over k of determinant 1 unit ball in a normed linear space X (norm) continuous dual of a nonmed linear space X

spaceX II

N(1)

(f:)

L2

discriminant of a basis x,,...,x discriminant ideal of a ring extension B/A left and right translation operators on f Haar covering number projective limit of a projective system {G,}

limG,

L3

Z

L3

Zo

L3

G° Gal(K/F)

1-3

FS

L3

IGI

L4

Aut(V)

algebraic automorphisms of a vector space

Auttop(V)

21 21

Hom(A,B)

2.2

bounded operators between Banach spaces

End(A)

22

endomorphisms on a Banach space A

11 TII

2.2

norm of a bounded operator T

L3

projective completion of Z ring ofp-adic integers connected component of the identity Galois group of the field extension K/F fixed field of a set S of automorphisms of F order of a profinite group G topological automorphisms of a topological vector space

Index of Notation

xvii

sp(a)

2.2

spectrum of an element in a Banach algebra

r(a)

2.2

spectral radius

A

space of characters of a Banach algebra A

d

22 22

TO(X)

23

T*

ATcEnd(H)

23 23

adjoint of an operator Tin a Hilbert space the closed, self-adjoins, unital subalgebra generated by Tin the ambient ring

T'n

23

Homc(V, V')

21

square root of a positive operator space of G-linear maps between two representation spaces

d

3.1

Xt")cG

3.1

W(K,V)

3_1

N(e)cS'

3_1

32 32 32

VV

f*g 3'(G)

Gelfand transform of a continuous functions that vanish at infinity

Pontryagin dual of G n-fold products within a group G local basis sets for the compact-open topology

a-neighborhood of the identity in S' Hilbert space associated with p convolution of functions continuous function of positive type, bounded by 1 on G

32 13

elementary functions on G

V(G)

33

complex span of continuous functions of positive type

V'(G)

3.3

Th

33

L'-functions in V(G) Fourier transform of a measure

mod0(a) Bmck

4.1

ordk(a)

42

r(G)

f

I

' ID. I

'I

;r=rtk

e=e(kilk)

4.1

42

42 43

Fourier transform of a function f

module of an automorphism a on G ball of module radius m in a topological field k order of an element of a local field k p-norm and infinity norm on Q or Fq(t); see also Section 42

uniformizing parameter for a local field k ramification index of an extension of local fields

xviii Index of Notation

f =f(k1Ik)

4.3

residual degree of an extension of local fields

K

4A 4.4

completion of a field K at a place v completion of global field K at the place corresponding to a prime Q

4.4 4A 4A 4A

set of places of K

KQ

'PKf rx/F:-'PK vIu o,,

4A 4.4

set of Archimedean places of K set of ultrametric places of K restriction map for places of a field extension K/F place v restricts to place u local ring of integers with respect to a place v

ring of integers of a global field K

DQ

4A 4.5

PQ

4.5

jQ

4.5

Homk(L,M)

4.5

embedding of L into M over k

II' G,,

51

GS

11

F1 dgy

I1

AK

5.2

IK

5-2

S.

5-2

A.

12

CK

5-4

IXI A,

54

restricted direct product S-version of the restricted direct product induced Haar measure on a restricted direct product of locally compact groups adele group of a global field K idele group of a global field K set of infinite places of a global field the open subgroup As, of the adele group idele class group of global field K; see also Section 6.4 standard absolute value on the adele group

CK =IK/K*

5.4

norm-one idele class group

S.

5.4

set of Archimedean places of a global field

IK S

I K.S

54 54

S-ideles of the global field K S-ideles of norm one

RS

5-4

AKS

5-A

S-integers of a global field S-adeles of the global field K

OK

.

decomposition group of a prime Q canonical map from DQ to Gal(Fq/F) induced isomorphism from DQ onto Gal(KQIFp) where Q lies over P

Index of Notation

xix

S-class group of a global field K discrete valuation associated with a prime P in a Dedekind domain

CK S

5A

VP

5-4

KAI,

£4

elements of K congruent to 1 modulo the integral ideal M

JA,(M)

54

C/A,(M)

1.4

KM

5A

fractional ideals relatively prime to M wide ray class group of K relative to M elements of K congruent to 1 modulo the ideal M extended by a set of real places

1

5A

narrow ray class group of K relative to M

&1

Frobenius element associated with primes Q and P, where Q lies over P

(P, K/F)

6 -1

Artin symbol (or Frobenius class)

F -(P)

6 -1

EF

6.2

(G,G)

(U

maximal unramified extension of Fat P set of places of a global field F commutator subgroup of a group G abelianization of a group G transfer map idele class group for F global, F* for F

CIK (M) 'PQIP

Gab

6

V:Gab->Hab

6

CF

6.4

local NK,F: CK -+ CF

f-4

norm homomorphism

JK,F : CF -> CK

6.4 6.4

map induced by inclusion Galois group of the separable closure of F over a finite extension K of F

6A 6A

inclusion map of Galois groups transfer map on Galois groups

(,4

Artin map

64

Artin map with projection onto Gal(K/F) maximal abelian extension of a field F extension of F by all nth roots of unity extension of F by all roots of unity theta function Haar measure on F* as given by the Haar measure dx on a local field F

rK= Gal F/K) i/C/F : rK -+ rF

V: rF -> rK gF:CF->rF

F

6 6

F..

6-5

0(z)

7A)

d*x= dx/I x I

21

OF

21 21

OXIF

F's

9F

elements of F* of unit absolute value valuation group of a local field F

xx

Index of Notation

sgn(x)

71 71 71

characters of a local field F local L-factor associated with a local character z; see also Section 74 ordinary gamma function gamma function associated with F= R or C sign character

XV =Z-11. 1

71

shifted dual of a character X

yv,

71

S(F)

Z(f,x)

71 71

of

U

multiplicative translate of an additive character by a field element a space of Schwartz-Bruhat functions on F local zeta function; see also Section 7.3 dual of of with respect to the trace map different of a field F standard character of a local field F Gauss sum for characters wand 2

X(F*) L(Z) or L(s,x)

F(s) FF(s)

DF WF

g(ra2) W(w) S(AK)

ip(x) Div(K) Div°(K)

71 71

71 71 71 11 72 72 2.2 7 22

div(J)

72 72

div(x)

2.2

Pic(K)

7-2

Pic°(K)

7 22

L(D)

2.2 72

deg(D)

yvK

73 73

root number associated with a character w adelic Schwartz-Bruhat functions average value of 97ES(AK) over K

divisor group of a function field K group of divisors of degree zero degree of a divisor D principal divisor associated with f divisor function extended to ideles; see also Section 7 5 Picard group of a function field K Picard group of degree zero linear system associated with a divisor D dimension of the vector space L(D) standard character of a global field K

Z(f Z)

7.3

local different at P of a global field global zeta function

L(s,Z)

14

Hecke L-function associated with a global

character x L(s,Z) L(s,Z.) c(s)

74 L4 74

finite version of L(s,Z)

infinite version of L(s,Z) Riemann zeta function

Index of Notation CA.(s)

74

reg(x)

7-5

dK

1-5

WK

7-5

RK

7.5

ri(K), r2(K)

7-6

S(S)

7-1

xxi

Dedekind zeta function regulator map discriminant of a number field K number of roots of unity in a global field K regulator of a number field number of real and nonconjugate complex embeddings of a number field K into C Dirichlet density of a set of primes S

1

Topological Groups

Our work begins with the development of a topological framework for the key elements of our subject. The first section introduces the category of topological groups and their fundamental properties. We treat, in particular, uniform continuity, separation properties, and quotient spaces. In the second section we narrow our focus to locally compact groups, which serve as the locale for the most important mathematical phenomena treated subsequently. We establish the essential deep feature of such groups: the existence and uniqueness of Haar measure; this is fundamental to the development of abstract harmonic analysis. The last two sections further specialize to profinite groups, giving a topological characterization, a structure theorem, and a set of results roughly analogous to the Sylow Theorems for finite groups. The prerequisites for this discussion will be found in almost any first-year graduate courses in algebra and analysis.

1.1 Basic Notions DEFINITION. A topological group is a group G (identity denoted e) together with a topology such that the following conditions hold: (i)

The group operation

GxG aG (g, h) -4 gh is a continuous mapping. (The domain has the product topology.)

(ii) The inversion map

G-G g H gis likewise continuous.

By convention, whenever we speak of a finite topological group, we intend the discrete topology.

1. Topological Groups

2

Clearly the class of topological groups together with continuous homomorphisms constitutes a category.

It follows at once that translation (on either side) by any given group element is a homeomorphism G-+G. Thus the topology is translation invariant in the sense that for all geG and UcG the following three assertions are equivalent:

(i) U is open. (ii) gU is open.

(iii) Ug is open. Moreover, since inversion is likewise a homeomorphism, U is open if and only if U-1 _ {x : x1 a U} is open.

A fundamental aspect of a topological group is its homogeneity. In general, ifX is any topological space, Homeo(X) denotes the set of all homeomorphisms X-+X. If S is a subset of Homeo(X), then one says that X is a homogeneous

space under S if for all x,ynX, there exists feS such that f(x)=y. (When S is unspecified or perhaps all of Homeo(K), one says simply that X is a homogeneous space.) Clearly any topological group G is homogeneous under itself in the sense that given any points g,hEG, the homeomorphism defined as left translation by hg-1 (i.e., x H hg lx) sends g to h. From this it follows at once that a local base at the identity eEG determines a local base at any point in G, and in consequence the entire topology. EXAMPLES

(1) Any group G is a topological group with respect to the discrete topology.

(2) R*, R'', and C* are topological groups with respect to ordinary multiplication and the Euclidean topology. (3)

R" and C" are topological groups with respect to vector addition and the Euclidean topology.

(4) Let k=R or C. Then the general linear group GL"(k) = (geM"(k) : det(g)#O) (nz 1) is a topological group with respect to matrix multiplication and the Euclidean topology. The special linear group

SL"(k) = {geGL"(k) : det(g)=1) (nZ 1) is a closed subgroup of GL"(k).

1. 1. Basic Notions

3

In subsequent discussion, if X is a topological space and xeX, we shall say that USX is a neighborhood of x if x lies in the interior of U (i.e., the largest open subset contained in U). Thus a neighborhood need not be open, and it makes sense to speak of a closed or compact neighborhood, as the case may be.

A subset S of G is called symmetric if S=S -1. This is a purely grouptheoretic concept that occurs in the following technical proposition.

1-1 PROPOSmoN. Let G be a topological group. Then the following assertions hold: (i) Every neighborhood U of the identity contains a neighborhood V of

the identity such that VVc U.

(ii) Every neighborhood U of the identity contains a symmetric neighborhood V of the identity. (iii) IfH is a subgroup of G, so is its closure. (iv) Every open subgroup of G is also closed. (v) If K1 and Kz are compact subsets of G, so is K, K2 .

PROOF. (i) Certainly we may assume that U is open. Consider the continuous map q: Ux U-+G defined by the group operation. Certainly-'(U) is open and contains the point (e, e). By definition of the topology on Ux U, there exist open subsets V,, V2 of U such that (e, e) a V, x VZ . Set V= V, n VZ . Then V is a neighborhood of e contained in U such that by construction VVg U.

(ii) Clearly g e U n U-' a g, g-' e U, so V= Un U-1 is the required symmetric neighborhood of e.

(iii) Any two points g and h in the closure of H may be exhibited as the limits of convergent nets in H itself. Hence by continuity their product is likewise the limit of a convergent net in H and similarly for inverses.

(iv) If H is any subgroup of G, then G is the disjoint union of the cosets of H, and hence H itself is the complement of the union of its nontrivial translates. If H is open, so are these translates, whence H is the complement of an open set and therefore closed.

(v) K,KZ is the image of the compact set K1xK2 under the continuous map (k,,k2) -> k,k2. It is therefore compact by general topology

Cl

Note that (i) and (ii) together imply that every neighborhood U of the identity contains a symmetric neighborhood V such that VVc U.

4


Translation of Functions and Uniform Continuity Given an arbitrary function f on a group, we define its left and right translates by the formulas ' Lhf(g)=f(h-`g) and R f(g) f(gh) .

If f is a (real- or complex-valued) continuous function on a topological group, we say that f is left uniformly continuous if for every a>0 there is a neighborhood V of e such that

heV= IILhf-f1.<e 11. denotes the uniform, or sup, norm. Right uniform continuity is defined similarly. Recall that ',(G) denotes the set of continuous functions on G with compact support. where 11

1-2 PROPOSITION. Let G be a topological group. Then every function f in W,(G) is both left and right uniformly continuous.

PROOF. We prove right uniform continuity. Let K=supp(f) and fix a>0. Then for every gEK there exists an open neighborhood U. of the identity such that h EUs

If(gh)-f(g)I<e

.

Equivalently, f(g') is -close tof(g) whenever g-Ig' lies in U. Moreover, by the comment following the previous proposition, each U. contains an open symmetric neighborhood s of the identity such that VBV, U9. Clearly the collection of subsets gV8 covers K, and a finite subcollection {g1V } ,, suffices. Henceforth we write VJ for Vg and U. for Ug . Define V, a symmetric open neighborhood of the identity e, y the formula 1

V=(,. J-t

If gEK, then gegll .. for some j. For hE V we consider the differencef(gh)-f(g):

If(gh) -f(g)I 5 If(gh) -f(g)I + If(g)-f(g)I The point is that both g 'g and g Igh lie in Uj, so that both terms on the right are bounded by & (Here is where we use that property VV,(z U, for all J.) This establishes right uniform continuity for K.

1.1. Basic Notions

5

When g does not lie in K, then we must bound I f(gh)I. If f(gh)#0, then ghe g,V for some j, and therefore f(gh) is e-close to AV. Moreover, gi-lg=gf ighlr' lies in Ui (here is where we use the symmetry of V), and it follows that If(g;)I<e since gi is close to g andf(g)=O by assumption. Consequently If(gh)I (ii) If G is T, then for any distinct g,heG there is an open neighborhood U of the identity lacking gh-1. According to Proposition 1-1, U admits a symmetric open subset V, also containing the identity, such that VVc U. Then Vg and Vh are disjoint open neighborhoods of g and h, since otherwise gh-1 lies

in V-'V=VVcU. (ii)=(iii) Every point in a Hausdorff (or merely T,) space is closed. (iii) =::> (iv) This is a consequence of homogeneity: For every point xeG there is a homeomorphism that carries e onto x. Hence if a is closed, so is every point. (iv) => (i) Obvious by general topology.

0

If H is a subgroup of the topological group G, then the set G/H of left cosets of G acquires the quotient topology, defined as the strongest topology such that the canonical projection p:g-4gH is continuous. Thus U is open in G/H if and only if p-'(U) is open in G. Recall from algebra that G/H constitutes a group under coset multiplication if and only if H is moreover normal in G. We shall see shortly that in this case G/H also constitutes a topological group with respect to t quotient topology.

6


The following two propositions summarize some of the most important properties of the quotient construction.

1-4 PROposrrlON. Let G be a topological group and let H be a subgroup of G. Then the following assertions hold: (1)

The quotient space GIII is homogeneous under G.

(ii) The canonical projection p: G-+GIII is an open map. (iii) The quotient space G/H is T, if and only if H is closed.

(iv) The quotient space G/H is discrete if and only if H is open. Moreover, if G is compact, then His open if and only if G/H is finite.

(v) If H is normal in G, then G/H is a topological group with respect to the quotient operation and the quotient topology.

(vi) Let H be the closure of (e) In G. Then H is normal in G, and the quotient group G/H is Hausdorff with respect to the quotient topology. PROOF. (i) An element xaG acts on G/H by left translation: xgH. The inverse map takes the same form, so to show that left translation is a homeomorphism of G/H, it suffices to show that left translation is an open mapping on the quotient space. Let U be an open subset of GIH. By definition of the quotient topology, the inverse image of U under p is an open subset U of G, and it follows that the inverse image of gUunder p is gU, also an open subset of G. Therefore gU is open, and left translation is indeed an open map, as required.

(ii) Let V be an open subset of G. We must show that p(V) is open in the quotient. But p(V) is open in G/H if and only if p '(p(V)) is open in G. By elementary group theory, Let x lie in so that x=vh for some ve V and h EH. Since V is open, given any ve V, there is an open neighborhood U c V containing v. Thus U,;h is an open neighborhood of x contained in which is accordingly open. (iii) By general topology, G/H is T, if and only if every point is closed. Since a

coset of H is its own inverse image under projection, each coset is a closed point in GIH if and only if each is likewise a closed subset of G. But by homogeneity this is the case if and only if H itself is closed in G. (Note that we cannot appeal to the previous proposition, since the topological space G/H is not necessarily a topological group with respect to multiplication of cosets.) (iv) Let H be a subgroup of G. Then by part (ii), H is an open subset of G if and

only if H is an open point of G/H. Since G/H is homogeneous under G, this

1. 1. Basic Notions

7

holds if and only if G/H is discrete. Assume now that G is compact. Then so is G/H, since p is continuous. But then H is open if and only if G/H is both compact and discrete, which is to say, if and only if G/H is finite. (Recall our convention that a finite topological group carries the discrete topology.) (v) Assume that H is a normal subgroup of G. Then from part (ii) and the commutative diagram

G

Ts

G/H

TO(s)

G yP

P

.

G/H

(where Tg denotes left translation by g), we see at once that translation by any group element is continuous on the quotient. A similar diagram establishes the continuity of the inversion map. (vi) Since {e} is a subgroup of G, so is its closure H. Moreover, it is the smallest closed subgroup of G containing e and therefore normal, since each conjugate of H is likewise a closed subgroup containing e. In light of the previous proposition, the full assertion now follows from parts (iii) and (v) above. Part (vi) shows that every topological group projects by a continuous homomorphism onto a topological group with Hausdorff topology. In this sense the assumption that a given group is Hausdorff is not too serious. 1-5 PROPOSIT[ON. Let G be a Hausdorff'topological group. Then the following assertions hold: (i) The product of a closed subset F and a compact subset K is closed.

(ii) If H is a compact subgroup of G. then p: G-*GIH is a closed map.

PROOF. (i) Let z lie in the closure of the product FK. Then there exists a net converging to z of the form (xaya) with and yaEK. Since K is compact, we may replace our given net by a subnet such that {ya} converges to some point y in K. We claim that this forces the convergence of {xa} in F to zy[, showing that z = zyly lies in FK, which is therefore closed. To establish the claim, consider an arbitrary open neighborhood U of the identity e. We may choose yet another neighborhood of e contained in U such that VVV U. Then the nets {z-'xaya} and {y.-'y) are both eventually in V, whence the product z-'xa ya ya' y = z--'x. y is eventually in U. Thus lim xa=zy-[, as required.

8


(ii) If X is a closed subset of G, then arguing as the second part of the previous proposition, we are reduced to showing that XH is likewise a closed subset of G. But if H is compact, this is just a special case of assertion (i). REMARK. The requirement that H be compact is essential. For example, in the case G=R2, with subgroup H={(O,y):yER}, we have clearly G/H=R, and under this identification, p(x,y) = x. Now let X= { (x, y) E R2 : xy = 1). Then X is closed, but p(X)=R* is not.

Locally Compact Groups Recall that a topological space is called locally compact if every point admits a compact neighborhood. DEFINITION. A topological group G that is both locally compact and Hausdorff is called a locally compact group.

Note well the assumption that a locally compact group is Hausdorff. Accordingly, all points are closed. 1-6 PROPOSITION. Let G be a Hausdorff topological group. Then a subgroup H of G that is locally compact (in the subspace topology) is moreover closed. In particular, every discrete subgroup of G is closed.

PROOF. Let K be a compact neighborhood of e in H. Then K is closed in H, since H is likewise Hausdorff, and therefore there exists a closed neighborhood U of a in G such that K= UnH. Since UnH is compact in H, it is also compact in G, and therefore also closed. By Proposition 1-1, part (i), there exists a

neighborhood V of e in G such that VVc U. We shall now show that

xEH )XEH.

First note that H is a subgroup of G by Proposition 1-1, part (iii). Thus if x r= H, then every neighborhood of x-' meets H. In particular, there exists some

yEVx-lnH. We claim that the product yx lies in UnH. Granting this, both y and yx lie in the subgroup H, whence so does x, as required. PROOF OF CLAIM. Since UnH is closed, it suffices to show that every neighborhood W of yx meets UnH. Since y-1 W is a neighborhood of x, so is y-1 WnxV.

Moreover, by assumption x lies in the closure of H, so there exists some element zey-' Wr- xVnH. Now consider: (i) the product yz lies both in Wand in the subgroup H;

(ii) by construction, ye Vx-'; (iii) by construction, zexV.

1.2. Haar Measure

9

The upshot is that yz lies in Vx-1 xV=VV, a subset of U, and therefore the intersection Wn(UnH) is nonempty. This establishes the claim and thus comU pletes the proof.

1.2 Haar Measure We first recall a sequence of fundamental definitions from analysis that culminate in the definition of a Haar measure. We shall then establish both its existence and uniqueness for locally compact groups.

A collection `m of subsets of a set X is called a a=algebra if it satisfies the following conditions:

(i) XETZ. (ii) IfAe`x12, thenAc&1R, where Ac denotes the complement ofA in X. (iii) Suppose that A,, r=T? (n 2- 1), and let 0

A=UA,. n=1

Then also A eÀ2; that is, T? is closed under countable unions.

It follows from these axioms that the empty set is in `:fit and that `m is closed under finite and countably infinite intersections. A set X together with a a.algebra of subsets `l)t is called a measurable space. If X is moreover a topological space, we may consider the smallest a6algebra . containing all of the open sets of X. The elements of .0 are called the Bore! subsets of X.

A positive measure p on an arbitrary measurable space (X T Z) is a function p:`]R-+R+.4ao) that is countably additive; that is,

P(U An)_Yp(An) for any family {An} of disjoint sets in V. In particular, a positive measure defined on the Borel sets of a locally compact Hausdorff space X is called a Borel measure. Let p be a Borel measure on a locally compact Hausdorff space X, and let E be a Borel subset off. We say that p is outer regular on E if

p(E) = inf{p(U) : UQE, U open)

.

10


We say that u is inner regular on E if

p(E) = sup(p(K) : KcE, K compact)

.

A Radon measure on X is a Borel measure that is finite on compact sets, outer regular on all Borel sets, and inner regular on all open sets. One can show that a Radon measure is, moreover, inner regular on a-finite sets (that is, countable unions of p-measurable sets of finite measure). Let G be a group and let p be a Borel measure on G. We say that p is left translation invariant if for all Borel subsets E of G,

p(sE) = p(E) for all seG. Right translation invariance is defined similarly. DEFINMON. Let G be a locally compact topological group. Then a left (respec-

tively, right) Haar measure on G is a nonzero Radon measure p on G that is left (respectively, right) translation-invariant. A bi-invariant Haar measure is a nonzero Radon measure that is both left and right invariant.

The following proposition shows that the existence of a left Haar measure is equivalent to the existence of a right Haar measure and, in a sense, equates the translation invariance of measure with that of integration. As usual, we let

W+(G)={f We often abbreviate this to 01' when the domain is clear.

1-7 PROPOSrrION. Let G be a locally compact group with nonzero Radon measure p. Then: (i) The measure p is a left Haar measure on G if and only if the measure

p defined by p(E) = p(E-1) is a right Haar measure on G. (ii) The measure u is a left Haar measure on G if and only if

f L, f dp = f f du o

c

for all fe 8.+ and se G.

(iii) If p is a left Haar measure on G, then p is positive on all nonempty open subsets of G and

1.2. Haar Measure

11

Jfdp>0

a

for allfE W,'.

(iv) If p is a left Haar measure on G, then p(G) is finite if and only if G is compact.

PRoOF. (i) By definition, we have the equivalence

p(E) = p(Es) Vs e G a p(E-') = As 'E-) Vs e G for all Bore] sets E; the assertion follows at once. (For any topological group G, clearly E is a Bore] subset of G if and only if E-' is.)

(ii) If p is a Haar measure on G, then the stated equality of integrals follows by

definition for all simple functions fe sio ` (i.e., finite linear combinations of characteristic functions on G), and hence, by taking limits, for arbitrary fe Z.. Conversely, from the positive linear functional J6 - du on ',(G) we can, by the Riesz representation theorem, explicitly recover the Radon measure p of any open subset Uc G as follows: P(U) = sup{ f f du : f e W.(G ),

II

./11.51, and supp(f) c U}

G

From this one sees at once that if the integral is left translation invariant, then

p(sU) = p(U) for all open subsets U of G, since supp(f) c U if and only if supp(L, f)csU. The result now extends to all Borel subsets of G because a Radon measure is by definition outer regular.

(iii) Since p is not identically 0, by inner regularity there is a compact set K such that p(K) is positive. Let U be any nonempty open subset of G. Then from the inclusion

KcUsU seo

we deduce that K is covered by a finite set of translates of U, all of which must

have equal measure. Thus since p(K) is positive, so is p(U). If fe Z,, then there exists a nonempty open subset U of G on which f exceeds some positive constant R. It then follows that

JfdpZRp(U)>0 G

as claimed.

12


(iv) If G is compact, then certainly u(G) is finite by definition of a Radon measure. To establish the converse, assume that G is not compact. Let K be a compact set whose interior contains e. Then no finite set of translates of K cov-

ers G (which would otherwise be compact), and there must exist an infinite sequence (ss) in G such that S. e U sjK j L,f (t) S EcjL. op(t) Vt E G which is to say that

f SIC'L"V) We have already seen that G is compact. Thus it remains to show that G'= (e). Let Ube any open subgroup of G. Then UnG° is open in G° and nonempty. Now consider the subset V of G defined by

V= SECT°-U

is open in G°, so is V. Moreover, by elementary Then since each group theory, U,-V=0, and G° is the disjoint union of two open sets, namely Un G ° and V. But by definition G ° is connected, so either Un G ° or V must be empty. Since the former is not, the latter is, and in fact G°= UnG°, which is to say that G° c U. Since U is an arbitrary open subgroup of G, we have accordingly,

G°c nU

.

U an open

sth n,of6 We must now make use of the profinite nature of G. Indeed, let

G=limG, where each G, is a finite group with the discrete topology. Recall that for each index i we have a projection map p.: G --* G. that is just the restriction of the corresponding map on the full direct product. Let y=(y) lie in G and assume that y is not the identity element. Then for some index io, it must be the case that y,° * e;o. But now consider the set U0 = p,- (e,). Since the topology on G,0 is discrete and the projections are continuous, Uo is open in G. Since the projections are moreover group homomorphisms, Uo is in fact a subgroup of G. But by construction, U0 excludes y. This shows that the only element in the intersection of all open subgroups of G is the identity. Thus G° is trivial, as required.

The proof of the converse is more delicate and requires three lemmas. We begin with some preliminary analysis.

Let,, be the family of open, normal subgroups of G. This is clearly a directed set with respect to the relation MsN if NcM. (In fact, two subgroups M and N in .41 have a least upper bound Mr-N in f.) Moreover, the following observations are elementary: (i) For each Near, , the quotient group GIN is both compact and discrete,

hence finite.

1.3. Profinite Groups

27

(ii) For each pair of subgroups M,NerV, with MSN, the kernel of the canonical projection G -> GIM contains N, and hence this map factors through GIN to yield the induced map

cMN:GIN -*GIM xN H xM . From this description it is clear that if LSMSN in A', then 97L.M° pM,N - c'L,N

and (GIN I.,,, constitutes a projective system of finite groups.

The point, of course, is to show that G is isomorphic to the projective limit of this system. 1-15 LEMMA. Let the profinite group G' be given by

G'= lim GIN

rN

where N varies over.4; as defined above. Then there exists a surjective, continuous homomorphism a: G-a G'. PROOF. For Ne-,r let aN denote the canonical projection from G to GIN, which is surjective. Since GIN is homogeneous, we establish that aN is also continuous by noting that aN (eQ,N) = N, which by hypothesis is open in G. Arguing as in (ii) above, it is clear that whenever MSN in A', the following triangle is commutative: GIN

GIM Thus by the universal property of projective limits, we have a continuous homomorphism a : G -* G' such that a, = pN o a for all where pN denotes

projection from G' onto GIN, the component of the projective limit corresponding to N.

It remains to show that a is surjective. We claim that a has dense image in G'. Granting this, we conclude the argument as follows: Since G is compact

28


and G' is Hausdorff, the image of a is, moreover, closed in G. Thus Im(a), being dense, must be all of G', as required. To establish the claim we shall show that no open subset of G' is disjoint from Im(a). Consider the topology of G': this is generated by sets of the form pN (SN) , where S. is an arbitrary subset of GIN. Every open set in G' is thus expressible as a union of finite intersections of these pN (SN) . Such an intersection U consists of elements of the form (xN

where at most only finitely many of the coordinates are constrained to lie in some given proper subset of the corresponding quotient; the rest are unrestricted. Now suppose that the constrained coordinates correspond to the subgroups N...... N, and that

M=nNJ

.

J=1

Then given (xx)EG', the coordinates xxJ are all determined as images of the coordinate xM under the associated projection maps. Since is surjective, there is at least one element in such that a(t)M=xM, and conse-

quently t also satisfies a(t)N=xN for j=1,...,r. In particular, if (xN)CU, then certainly a(t)e U, since a(tlagreees with (xN) in all of the constrained coordinates. Thus U manifestly intersects Im(a), and by our previous remarks, so, too, does every open set in G'. This completes the proof. 0 1-16 LEMMA. Let X be a compact Hausdorff space. For a fixed point PC-X, set Wl = (K: K is a compact, open neighborhood of P). Define YcX by

Y= nK

.

KEW

Then Y is connected.

PROOF. Note that the collection `W is nonempty because X itself is compact and open.

Suppose that Y is the disjoint union of closed subsets Y, and Y2. We must show that either Y, or Y2 is empty. Recall from general topology that a compact

Hausdorff space is normal. Accordingly, there exist disjoint open subsets U, and U2 containing, respectively, Y, and Y2. Now set Z=X-(Uk-i U2), which is closed and therefore compact. Since Yc Ulv U2, Z and Y are disjoint, which is to say that Z lies in the complement of Y. Thus we have an open cover for Z


29

Z(-- UK` KEW

that admits a finite subcover. Hence there exist K...... K,E (such that J

Let W denote the intersection of the K. Then W is a compact, open neighborhood of P, and so W is itself in W. But also

W=(WnU,) U (WnU2) since W is disjoint from Z, the complement of U,v U2. We now make note of the following assertions: (i)

Both Wn Ui and Wn U2 are compact, open subsets of X.

(ii) P lies exclusively in one of Wr, Ul or Wn U2. Say PE Wn U,

From (i) and (ii) it follows that Wn U, ei! and so Yc Wn U,. Since Y2c Y and LI Y2 is disjoint from U, it follows that Y2 is empty, as required.

1-17 LEMMA. Let G be a compact, totally disconnected topological group. Then every neighborhood of the identity contains an open normal subgroup.

PROOF. As a preliminary, note that G is Hausdorff: If x and y are distinct points in G, then {x,y} is disconnected with respect,to the subspace topology. Therefore there exist respective open neighborhoods of x and y that are disjoint. The proof now proceeds in three steps: First, we show that every open neighborhood U of the identity contains a compact, open neighborhood W of the identity. Second, we show that W in turn contains an open, symmetric neighborhood V of the identity such that WVc W. Third, from V we construct an open subgroup, then an open, normal subgroup of G contained in U, as required. Let Y_/ denote the set of all compact, open neighborhoods of the group identity e. Applying the previous lemma with P=e, we find that

Y= nK Keg!

is a connected set containing e. But G is totally disconnected, so in fact Y={e}.

Now let U denote any open neighborhood of e. Then G- U is closed and therefore compact. Since e is the only element of G common to all of the K in

30


W,, there exist subsets Ks,...,Kre?l whose complements cover G-U, and therefore r

w=f1K; J=1

is a subset of U and a compact, open neighborhood of e. In particular, Well. This completes the first step. To begin the second step, consider the continuous map p:WxW-'G defined by restriction of the group operation. We make the following observations:

(i) For every we W, the point (w,e)e u-'(W). (ii) Since W is open, the inverse image of W itself under u is open in Wx W.

(iii) It follows from (i) and (ii) that for every we W, there exists open neighborhoods U,, of w and V of e such that Uwx V cu-' (W). Moreover, by Proposition 1-1, we may assume that each ,,, is symmetric. (iv) The collection of subsets UK, (we W) constitutes an open cover for W. Since W is compact, a finite subcollection U,'...' Ur suffices.

Let V1,..., V correspond to U...... Ur in (iii) above. Define an open neighborhood Vc W of the identity as follows:

V=flv, J1

By construction WVc W, and by induction WV"(-- W for all nZO. In particular, V"c W for all n Z O. This completes the second step. For the final step, we expand V to an open subgroup 0 of G contained in W by the formula

O=UV"

.

fit

(Note that 0 is closed under inversion because V is symmetric.) The quotient space GIO is compact and discrete, hence finite, so we can find a finite collection of coset representatives x1,...,x5 for 0 in G. It follows that 0 likewise has only finitely many conjugates in G: all take the form x,Ox'-' Thus

(j=l,...,s).


31

N=n"Ox;' J=J

is an open, normal subgroup of G. Moreover, since one of the conjugates of 0 is O 0 itself, NcOcWc U. This completes the proof.

This brings us at last to the conclusion of the topological characterization of profinite groups. PROOF OF THEOREM 1-14, CONVERSE. By Lemma 1-15, we have a surjective

homomorphism a: G-->G', where G' is the projective limit of the finite quotients GIN for N an open, normal subgroup of G (i.e., Ne ..4'). Appealing to Exercise 9 below, we see that it suffices to show that a has trivial kernel and hence is injective. Since a simultaneously projects on all of the quotients, it is clear that

Ker(a) = n N . Ne..Y

By the previous lemma, every open neighborhood of eeG contains an open, normal subgroup, which is therefore represented in the intersection above. It follows that Ker(a) is contained in every neighborhood of e and hence in the intersection of all such neighborhoods. But G is Hausdorff: the intersection of all neighborhoods of e consists merely of e itself. Hence Ker(a) is indeed trivO ial, and the theorem is proved.

The Structure of Profinite Groups The following theorem shows in particular that closed subgroups of profinite groups and profinite quotients by closed normal subgroups are likewise profinite.

1-18 THEOREM. Let G be a profrnite group and let H be a subgroup of G. Then His open if and only if G/H is finite. Moreover, the following three statements are equivalent. (i)

His closed.

(ii) His profrnite.

(iii) H is the intersection of a family of open subgroups.

Finally, if (i)-(iii) are satisfied, then GIH is compact and totally disconnected.

32


PROOF. The first statement follows from Proposition 1-4, part (iv), since a profinite group is necessarily compact. We next establish the given equivalences.

(i)=(ii) H is a closed subset of a compact space and therefore itself compact. Hence it remains to show that H is totally disconnected. But this is trivial: since G°={e}, also H°={e}, and this suffices by homogeneity. (ii)=(i) If H is itself profinite, it is a compact subset of a Hausdorff space and hence closed.

(iii)=>(i) Suppose that H is the intersection of some family of open subgroups of G. Then since every open subgroup is also closed [Proposition 1-1, part (iv)], H is also the intersection of a family of closed subgroups of G, and therefore itself closed.

(i)=(iii) As above, let .i{''denote the family of all open, normal subgroups of G. If Ne.iV, then since N is normal, NH is a subgroup of G. By part (i), [G:N] is finite, whence [G: NH] is likewise finite and NH is open. Moreover, clearly

FI c n Aril, NEA'

It remains only to demonstrate the opposite inclusion. So let x lie in the indicated intersection, and let U be any neighborhood of x. Then Ux-I is a neighborhood of e, and so by Lemma 1-16, Ux-' contains some Noe.iV. Since x lies in the given intersection, xeN0H. Now by construction, also xENox. Hence Nox is equal to Noh for some heH, and consequently hENoxe U. The upshot is that every neighborhood of x intersects H, and hence x lies in the closure of H. But H is closed by hypothesis, and therefore XEH, as required. For the final statement, the compactness of the quotient follows at once from

the compactness of G. Let p:G- *G/H denote the canonical map. To see that G/H is totally disconnected, assume that p(X) is a connected subset of Gill that properly contains p(H). Then Y=X-H is nonempty, and since we may assume that H is nontrivial, Y contains more than one point. Hence Y is the disjoint union of nonempty open (hence closed) sets F, and F2. One checks easily that since H is closed, F, and F2 are both open (hence closed) in X. Thus X is the disjoint union of the two nonempty closed sets F, vH and F2. But then the image of F2 under p is (a) nonempty, (b) not the full image of X, and (c) both open and closed in p(X). Since p(X) is connected, this is a contradiction. Hence the connected component of p(H) is p(H) itself, and the quotient is totally disconnected, as claimed.


33

A Little Galois Theory We close this section by showing how profinite groups make a momentous appearance in connection with the Galois theory of infinite extensions. To begin, we recall the following elements of field theory:

Let F be a field. An element a that is algebraic over F is called separable if the irreducible polynomial of a over F has no repeated roots. An algebraic field extension K/F is called separable if every element of K is separable over F. (ii) Assume that K is an algebraic extension of F contained in an algebraic closure F of F. Then we call K/F a normal extension if every embedding of K into F that restricts to the identity on F is in fact an automorphism of K. (We say that such an automorphism is an automorphism of K over F.) (iii) A field extension K/F is called a Galois extension if it is both separable and normal. The set of all automorphisms of K over F constitutes a group under composition; this is called the Galois group of K over F and denoted Gal(K/F). If FcLQK is a tower of fields and K/F is Galois, then K/L is likewise Galois. (i)

Note that these notions do not require that K/F be finite. Our aim now is to extend the fundamental theorem of Galois theory to infinite extensions. This will require the introduction of some topology.

If S is any set of automorphisms of a field F, as usual Fs denotes the fixed field of S in F; that is, the subfield of F consisting of all elements of F left fixed by every automorphism of S. Suppose that K/F is a Galois extension with Galois group G. Consider the set Y of normal subgroups of G of finite index. If N, Me./rand MQN, we have a projection map p,,,: G/M-+G/N, and hence a projective system of quotients {G/N},,s,r. This system is certainly compatible with the family of canonical

projections pN : G --* G/N, which corresponds to the restriction map from Gal(K/F) to Gal(KN/F). Thus we have a canonically induced homomorphism p from G into the projective limit of the associated quotients. 1-19 PROPOSITION. Let K, F, G, and Vbe as above. Then the canonical map

p:G-- lam GIN is in fact an isomorphism of groups. Hence G is a profrnite group in the topology induced by p.

34


In this context, we shall simply speak of the Galois group G as having the profinite topology. PROOF. We show first that p is injective. Certainly

Ker(p) = n N Ne.r

and so we need only demonstrate that this intersection is trivial. Let oeKer(p) and let XEK. Then by elementary field theory there exists a finite Galois exten-

sion FIF such that FcK and xEF. Now the restriction map from G=Gal(KIF) to Gal(F/F) has kernel Gal(K/F), which is therefore a normal subgroup of G of finite index. But then aEGal(K/F), and so o(x)=x. Since x is arbitrary, a is the identity on K, and Ker(p) is trivial, as required. We show next that p is also surjective. Fix (QH) in the projective limit. Given an arbitrary element xeK, again we know that x lies in some finite Galois ex-

tension P of F with N=Gal(KIF') normal and of finite index in G and Gal(F/F)=GIN. Now define QEGal(K/F) by a(x)=a,(x). By construction of the projective limit, o is independent of the choice of extension F, and hence is a well defined automorphism of K. Moreover, it is clear that or is j. (a) for all N.

Note that the isomorphism constructed in the previous proposition is essentially field-theoretic, and not merely group-theoretic. (See Exercise 12 below.) 1-20 THEOREM. (The Fundamental Theorem of Galois Theory) Let K/F be a

Galois extension (not necessarily finite) and let G=GaI(K/F) with the profinite topology. Then the maps

a : L H H = Gal (K/L)

f:HHL=KH constitute a mutually inverse pair of order-reversing bijections between the set of intermediate fields L lying between K and F, and the set of closed subgroups of G. Moreover, L is Galois over F if and only if the corresponding subgroup H is normal in G.

PROOF. Note that in the case of a finite extension KIF, we may ignore the topological restriction, and the statement amounts to the fundamental theorem of Galois theory for finite extensions, a result that we assume. We proceed in four steps.

STEP 1. We must show first that the map a is well-defined; that is, that a indeed yields closed subgroups of G. (The map O is of course well-defined on ar-


35

bitrary subsets of G.) According to the previous proposition, H is profinite as the Galois group of KIL, and Exercise 14 shows that this topology is identical to that induced by G. Thus H is a profinite subgroup of a profinite group and is therefore closed by Theorem 1-18. a is the identity map. Let L be an intermediate field. By definition a(L) fixes L, and so clearly 8(a(L));2L. Conversely, suppose that z lies in fl(a(L)). Then since z lies in K and is therefore separable over L, z also belongs to a finite Galois extension M of L contained in K. Let aeGal(M/L). STEP 2. We claim that

Then there exists aEGal(K/L) that restricts to a. (The extensibility of automorphisms for infinite extensions follows from the finite case by Zorn's lemma.) By construction, a(z) = z, and hence a(z) = z for all aeGal(M/L). But by the fundamental theorem for finite extensions, we know that zEL. Hence we have also that 8(a(L))cL, and the claim is established. STEP 3. We shall show now that ao# is likewise the identity. By definition, for any subgroup H of G we have that a(Q(H));;?H. Now assume that H is closed. Then again by Theorem 1-18, H is the intersection of a family Il of open subgroups of G. Since a and # are clearly order reversing,

rg(H)=Q( fu)a UQ(w) Uew

Uew

and

a(ft(H))ca( U/.3(U))c na(fi(U))= nU=H Ue9!

MEW

.

UeWW

The point is that each of the open subgroups U has finite index, and thus in each case a(Q(U))= U by the finite theory. STEP 4. Finally, suppose that a(L)=Gal(K/L)=H, where L is some intermediate field. Let a lie in G. Then from the diagram

K

L

0

Cr

F

K

o(L)

36


we deduce that Gal(K/o(L))= aH&-'. Thus according to parts (i)-(iii) above, we have that a(L)=L for all auG if and only if aHd-l=H for all aEG. This is to say that L is normal (and hence Galois) over F if and only if H is normal in G.

REMARK. We leave it to the reader to determine the effect of aoi4 on an arbitrary subgroup of Gal(K/F). (See Exercise 15 below.)

1.4 Pro-p-Groups Our aim here is to introduce for profinite groups an analogue of the p-Sylow subgroups that play such a crucial role in finite group theory. To begin, we must first generalize the notion of order.

Orders of Profinite Groups DEFINITION. A supernatural number is a formal product rip Ô P

where p runs over the set of rational primes and each nEENv (oo) .

Clearly the set of supernatural numbers is a commutative monoid with respect to the obvious product. If a is a supernatural number, we set vv(a) equal to

the exponent of p occurring in a. We say that a divides b, and as usual write alb, if vP(a) 5 vP(b) for all primes p. Note that if a I b, there exists a supernatural number c such that ac=b.

Given supernatural numbers a and b, we may define both their least common multiple and greatest common divisor by the formulas

lcm(a, b) _ f pap(v°(a) v°(b)) and gcd(a, b) _ f One extends these notions to arbitrary (even) infinite families of supernatural numbers in the obvious way.

Now let G be a profinite group. As previously, let A' denote the set of all open, normal subgroups of G. Recall that each quotient group GIN, for NeA', is finite. DEFINITION. Let H be a closed subgroup of G. Then we define [G: H], the index of H in G, by the formula

1.4. Pro-p-Groups

37

[G: HJ = 1cm [G/N: HN/N] . NE.r In particular, [G:{e}], the index of the trivial subgroup, is called the order of G and denoted IGI. Using the standard isomorphism between HN/N and H/HnN, we may recast the definition above as

[G:H]= 1cm [G/N:HIHnN]. Nc.#

See also Exercise 16 below.

1-21 PROPOSITION. Let G be a profinite group with closed subgroups H and K

such that HcK Then [G:Kj=[G:H][H:K]. PROOF. Note that since H is closed, it is also profinite, and so the assertion is well defined. Now let N be any open normal subgroup of G. Then

[GIN:K/KnNJ = [G/N:H/HnN] [H/HnN:K/KnN] .

(1.3)

The 1cm (over Ne.41) of either side of the equation is, of course, [G:HJ. Consider the factors on the right: if we replace N by any smaller subgroup N, C=_4" both indices are inflated (cf. Exercise 17). Hence, taking intersections, any pair of prime powers occurring in [G/N:H/HnN] and [H/HnN:K/KnN], respectively, may be assumed to occur simultaneously. The upshot is that we can compute the 1cm of the product by separately computing the Icm's of each factor. The first yields [G: HI; it remains only to show that the second yields [H:K]. Let M be any open, normal subgroup of H. Then M=Hn U, where U is open in G. But by Lemma 1-17, U contains an open, normal subgroup N of G, and one argues as above that

[HIM:K/KnM] I [H/HnN:K/KnN] . Thus [H:K) may be computed as the 1cm over subgroups ofHof the form HnN,

where N is open and normal in G. Hence the second factor on the right of O

Eq. 1.3 indeed yields [H: K], as required.

REMARK. The proof shows that we may compute a profinite index as the icm over any cofinal family rl of open normal subgroups of the ambient group; that is, if for every NE.#'there exists an Me.4" such that McN, then

1cm [G/N:HN/NJ = 1cm [GIM:HMIMj Ne.r ME!

.

38


EXAMPLES

(1) Consider the p-adic integers

Z. = lim(Z/p"Z) . Let if,, denote the kernel of the projection map from Zo to Z / p"Z. Since and it follows that p°° this projection is surjective, we have divides I Zp I. Conversely, every finite quotient of Zp has order a power of p, and therefore I ZD I =p'.

(2) Next consider

Z = lim(Z/nZ) n21

Arguing as above, every factor group Z/nZ occurs as a quotient of Z, whence every positive integer is a divisor of its order. Thus

IZI = Hp-

.

p prime

Pro-p-Groups

Let p be a rational prime. Recall that a group is called a p-group if the order of every element is finite and a power of p. In the case that G is finite, this is equivalent to the statement that the order of G is a power of p. DEFTNTrION. A projective limit of finite p-groups is called a pro -p-group.

Of course, Z, is a pro -p-group; so is Hp, the projective limit of the Heisenberg groups H(Z/p"Z). (See Exercise 18 below.) 1-22 PROPOSITION. A profinite group G is a pro -p-group if and only (fits order is a power of p (possibly Infinite).

PROOF. G) We have already seen in the proof of Theorem 1-14 that G is the projective limit of its finite quotient groups GIN. If the order of G is a power of p, then each of these quotients must be a p-group, as required.

=*) Suppose that G is the projective limit of the projective system P, of pgroups. Then by definition of the topology of G, cofinal among the open normal subgroups of G are subgroups of the form

1.4. Pro-p-Groups

39

M=(fQ;)nG where Q,=P; for all but finitely many indices, and Q,={ef} for the exceptions. Now given an arbitrary xEG and specifying any finite subset of its coordinates, there is clearly a finite exponent of the form q=p' such that x4 is trivial at each of the specified coordinates. Hence GIM is a p-group, and it follows by the remark following Proposition 1-21 that the order of G is a power of p.

DEFINrnON. Let G be a profinite group. A maximal pro-p-subgroup of G is called a pro-p-Sylow subgroup of G (or more simply, a p-Sylow subgroup of G).

Note that the trivial subgroup may well be a pro-p-subgroup of G for some primes p. The following theorem shows among other things that this is the case if and only ifp does not divide the order of G. 1-23 THEOREM. Let G be a profinite group and let p be a rational prime. Then the following assertions hold: (i)

p-Sylow subgroups of G exist.

(ii) Any pair of conjugate p-Sylow subgroups of G are conjugate. (iii) IfP is a p-Sylow subgroup of G, then [G: P] is prime to p. (iv) Each p-Sylow subgroup of G is nontrivial if and only if p divides the order of G. PROOF. As usual, let

denote the set of open normal subgroups of G and recall the explicit isomorphism

o:G -+lim GIN

r

x i-i (xN),,E,r

.

Note in particular that if x,yeG and xN=yN for every open normal subgroup N, then x=y. A similar statement holds for arbitrary subsets of G.

(i) For each Ne.d', let .4a(N) denote the set of p-Sylow subgroups of the finite group GIN. Then clearly .9(N) is finite and, moreover, nonempty. (If GIN has order prime top, then the trivial subgroup is a p-Sylow subgroup.) Assume that M,NFiYwith NcM. Then there exists a surjective homomorphism of finite groups N:GIN->GIM. Since this map sends a p-Sylow subgroup of GIN to a p-Sylow subgroup of GIM (refer again to Exercise 17), we obtain an induced map N:.41(N)->.'(M). Thus we obtain a projective system (9a(N), 97.. N) of finite nonempty sets, and the projective limit of this system is likewise nonempty by Proposition 1-11. This means that there exists a projective system of

40


where for each NE.M, we have PN c G/N. Let P p-Sylow subgroups be the projective limit of the PN, which we can clearly identify with a subgroup of the projective limit of the GIN and hence with a subgroup of G via qD. Then P is a pro -p-group by construction, and we shall now show that it is maximal. Let Q be any prop-subgroup containing P. Then for every open normal subgroup N, QNIN;2PNIN=PN. But Q is a pro-p-group, so by the previous proposition, QNIN is a p-group and therefore equal to the p-Sylow subgroup PN. Thus for every open normal subgroup N, QNIN=PNIN, and therefore Q and P have the same image under p and accordingly are equal. Hence P is indeed maximal, as claimed.

(ii) Let P and Q be p-Sylow subgroups of G. For every NEA', we make the following definitions:

PN = PNIN QN = QN/N YN = {yNEGIN : yNPNyN = QN}

Note that each YN is finite and, by the Sylow theorems for finite groups, nonempty. Moreover, the subsets YN clearly constitute a projective system. Let Y denote the (nonempty) projective limit of the YN, which we again identify with

a subset of G via to and let y lie in Y. Then by construction, yPy-' and Q have equal projection in GIN for all open, normal N and are therefore equal. Hence P and Q are indeed conjugate. (iii) Let P be a p-Sylow subgroup of G. Then by definition

[G: PI = lcm [GIN:PNIN] . 1V CX

But by Exercise 19, for each N, the subquotient PNIN is a p-Sylow subgroup of GIN, and so by finite group theory each index [GIN: PNIN] is prime to p. Hence [G: P] is likewise prime top. (iv) This follows at once from parts (i) and (iii). 1-24 COROLLARY. Let G be a commutative profinite group. Then the following assertions hold..-

(i)

For every prime p, G admits a unique pro p-Sylow subgroup.

(ii) Let p and q be distinct primes and let P and Q be the corresponding

Sylow subgroups. Then PnQ Is trivial. (iii) G is isomorphic to the direct product of its Sylow subgroups.

1.4. Prop-Groups

41

PROOF. (i) In light of the commutativity of G, this follows at once from parts (i) and (ii) of the theorem above.

(ii) The order of PnQ must divide powers of both p and q, whence this intersection must be trivial.

(iii) Let N be an open normal subgroup of G. Then for each pro-p-Sylow subgroup P we have a canonical projection from P onto PNIN, the unique p-Sylow subgroup of GIN. Note that this projection is trivial for all but the finitely many primes p that divide the order of GIN. By the theory of finite commutative groups, we have

fl PN/N - GIN where the product is taken over all of the Sylow subgroups of G. We may lift this isomorphism to G as follows:

G = lim GIN 4-

= lim fl PNIN = H lim PN/N

=fllim P/PnN =J JP

.

All products are over the set of Sylow subgroups of G; all projective limits are over the family of open, normal subgroups of G. The final line of the calculation is justified by the cofinality of subgroups of the form PnN among the open subgroups of P, which may be deduced from Lemma 1-17. 0 EXAMPLE. Recall that the abelian profinite group

Z = lim Z/nZ has order np°', where the product is taken over all primes. Given a prime p, let

P be the unique corresponding p-Sylow subgroup of Z. Let P. denote the unique p-Sylow subgroup of Z/nZ. Then

P = lim P = Urn Z/ p°°(")Z = lim Z/ pMZ = Z P 4.. n

Fn

Thus according to the corollary, Z = n Zr

M

.

.

42


Exercises 1.

Let G be a topological group. Show that the topology on G is completely determined by a system of open neighborhoods of the identity e.

2.

Let G=Z and impose the following topology: USG is open if either O U or G- U is finite. Show that G is not a topological group with respect to this topology. [Hint: If so, the mapping a H a + 1 would be a homeomorphism; show that it is not.]

3.

This exercise shows that we may impose a nondiscrete topology on Z such that Z is nonetheless a topological group with respect to addition. Let S' denote the multiplicative group of complex numbers of absolute value 1. Recall that an element of Hom(Z,S) is called a character of Z. We denote such a character X. Let 9?_1ISi X

where the product is taken over all characters. Then W is a compact topological group. Now consider the homomorphism

j:Z -+ n H (X(n)) (a) Show that j is injective; that is, show that for any nonzero n e Z there exists a character X such that X(n) * 1.

(b) Let G=j(Z). Then G is a group algebraically isomorphic to Z and a topological group with respect to the subspace topology induced by 9?. Show that G is not discrete with respect to this topology and conclude that Z itself admits a nondiscrete topological group structure with respect to addition. [Hint: Suppose that j(1) is open. Then there exists an open subset U of .? such that Un G =j(l); moreover, we may assume that all but finitely many projections of U onto its various coordinates yield all of S'. Noting that j(l) generates the infinite group G, one may now derive a contradiction.] 4.

Give an example of a topological group with a closed subgroup that is not open.

5.

Let X be a topological space and let C(X) denote the space of connected components of X. (This constitutes a partition of X). As usual, we impose

Exercises

43

the quotient topology on C(A)--the strongest topology such that the canonical projection p:X-*C(X) is continuous. Show that C(X) is totally discon-

nected with respect to this topology. [Hint: We say that a subset Y of a topological space is saturated if whenever yeY, the entire connected component of y lies in 1'. Let F be a connected component of C(X) that contains more than one point. Show that P -'(F) is a saturated, closed, disconnected set. Write p-'(F) as the disjoint union of two saturated, closed subsets of X, and apply p to this decomposition to show that F is in fact disconnected-a contradiction.] 6.

Let G=GL,,(R). Show that G° is the set of nxn matrices with positive determinant.

7.

Let H be a subgroup of the topological group G. Show that its closure H is normal (respectively, abelian) if H is.

8.

Let f : G --> G' be a surjective continuous homomorphism of topological groups. Show that f factors uniquely through G/Ker(J); that is, there exists a unique continuous homomorphism f such that the following diagram commutes:

G

f

G'

G/Ker(f) Show that f is moreover injective. Under what conditions is f a topological isomorphism onto its image? 9.

Let f : X -. Y be a continuous bijective mapping of topological spaces and assume that X is compact and Y is Hausdorff. Show that f is moreover a homeomorphism. [Hint: It suffices to show that f is open. What can one say about the image of U` under! where U is any open subset of X?]

10. Let I be an index set with preordering defined by equality and let (G;, 4pii) be a projective system of sets defined with respect to I. What is the projective limit in this case? 11. Give an example of a projective system of finite nonempty sets over a pre-

ordered, but not directed, set of indices such that the projective limit is nevertheless itself empty.

12. Let G be an arbitrary group. Show that in general G is not isomorphic to the projective limit of the quotient groups GIN, as N varies over all of the

44


subgroups of G of finite index. Hence not every abstract group acquires a profinite structure by this device. [Hint: Take G=Z.J 13. Let (G,, q',) and (H,, g,,) be two projective systems of sets. (Note that we use

the same map designators q' for both systems.) Suppose that we have a family of maps (C,:G,-sH,} that is compatible with these systems in the sense that q o C= r,o qy for all pairs of indices i sj. Show that there exists a

unique map ;'G-+H on their respective projective limits such that 4 o p,=p, o C for all i, where p, denotes the appropriate projection map. Observe that this construction works equally well in the categories of groups, topological spaces, and topological groups. [Hint: In light of the universal

property of projective limits, consider the family of composed maps {op,: G-*H,}.J 14. Let KIFbe a Galois extension with Galois group G.

(a) Let L be an intermediate field that is finite over F. For any given QeG, define NL(a)cG to be the set of reG such that o and r agree on L. The subsets NL(o) constitute a subbase for a topology on G. Show (i) that this topology remains unchanged if we restrict the subbase to normal intermediate fields that are finite over F and (ii) that this topology is identical to the profinite topology on G. (b)

Now let L be an arbitrary intermediate field, and let H denote the Galois group of K over L. Use the characterization of the profinite topology given in part (a) to show that the topology induced on H by G is identical to the profinite topology defined directly on H as Gal(K/L).

15. Let K/F be a Galois extension (not necessarily finite) and let H be any subgroup of G=Gal(K/F) (not necessarily closed). Let a and fl be defined as in Theorem 1-20. Show that a(fl(H)) = 9, the closure of H. 16. Let G be a profinite group and let H be a closed subgroup. Show that [G: H]= 1cm [G:HNJ

where ..V is the set of all open, normal subgroups of G. Show further that if M is any open subgroup of G containing H, then there exists an open normal subgroup N of G such that MQNH. Conclude from this and the previous equation that moreover, [G: H] = Icm [G : MJ M open MAN

Exercises

45

17. Let V : G -a G' be a suoective homomorphism of groups with kernel L. Let H be a subgroup of G of finite index and let H' be the image of H under op.

Show that [G:H]=[G':HI-[HL:H]. 18. For any commutative ring A with unity, define the Heisenberg group H(A) over A by

H(A) =

1

a

c

0

1

b

0

0

1

: a,b,c EA

(a) Show that H(A) is a group under multiplication in the matrix ring M3(A) and that this construction is, moreover, functorial in A. To continue, for n 2t 1, H(Z/p"Z) is clearly a group of order p 3n, and hence a p-group. If m In, then by functoriality, we have that the canonical projec-

tion Z/p"Z-o.Z/p"'Z induces a homomorphism op. from H(Z/p"Z) to H(Z/p"'Z). (b) Show that (H(Z/p"Z),q,,,,,) is a projective system of groups.

Let if, denote the projective limit of the H(Z/p"Z); by definition, this is a pro-p-group.

(c) Show that H(Z,) . Hl. [Hint: Consider the map

,r":H(Z,,) - H(Z/p"Z) induced by projection from Z onto Z/p"Z. Show that this is a continuous surjective homomorphism anuthat moreover, the family (n,,) is compatible with the system of homomorphisms Finally, show that the map ar obtained from the ;" by the universal property of the direct limit is the desired isomorphism.] 19. Let G be a profinite group and p a rational prime. For each open, normal subgroup N in G, let HN be a p-subgroup of GIN (not necessarily a p-Sylow subgroup). Show that there exists a pro-p-Sylow subgroup P of G such that

PN/N2HN for all N. Conclude (i) that every pro-p-subgroup of G is contained in a pro-p-Sylow subgroup of G; and (ii) that if P is a pro-p-Sylow subgroup of G, then PNIN is a p-Sylow subgroup of GIN for each open, normal subgroup N of G. [Hint: Generalize the argument from the proof of part (i) of Theorem 1-23.1

2 Some Representation Theory

The general background for Tate's thesis involves locally compact groups, their representations, and duality theory. Many of these basic prerequisites are derived in this and the next chapter.

Here we develop elements of representation theory for a locally compact topological group G represented in the automorphism group of a topological vector space V. A representation in this context is in fact a restricted instance of an ordinary abstract group representation, with the extra constraints involving continuity and some specific topological conditions on V. Our development is somewhat general without becoming excessively technical; in particular, we postpone the assumption that G is commutative until as late as possible. This is not empty abstraction: the noncommutative case is interesting in its own right, as shown by Jacquet-Langlands theory, which deals with representations of the general linear group. The key results of this chapter are Schur's lemma for irreducible unitary representations of a topological group G and the theorem that such representations are one-dimensional in the case that G is abelian. Considering that the finite-dimensional analogues of these statements are not particularly deep, they are surprisingly challenging to prove. In fact, the chase will lead us through the spectral theory of Banach algebras, the Gelfand transform, and the spectral theorems. (We state the second spectral theorem for completeness, but make no essential use of it.) The Gelfand transform is especially noteworthy because it is applied again in the following chapter in a wholly different context.

2.1 Representations of Locally Compact Groups A field k (subject to some given topology) is called a topological field if both addition and multiplication are continuous functions on kxk. A vector space V (again subject to some given topology) over k is called a topological vector space if the following two conditions are satisfied: (i) The underlying additive group (V,+) is moreover a topological group.

(ii) The scalar multiplication map

2.1. Representations of Locally Compact Groups

47

kxV --4V (A, v) ra AV

is continuous (with respect to the product topology on kx V). EXAMPLES

(1) If k is a topological field and V is any merely algebraic vector space over k, then we have an isomorphism of vector spaces

V=rl k I

where I is some index set. We may use the isomorphism to transfer the product topology of n k to V. One checks easily that with respect to this induced topology, V is a topological vector space over k. Moreover, for finite-dimensional V, every linear map is clearly continuous, and hence the transferred topology is independent of the choice of isomorphism.

(2) Recall that a normed vector space V over R (respectively, over C) that is complete with respect to the norm metric is called a real (respectively, complex) Banach space. One checks easily that V is a topological vector space over R (respectively, C) with respect to the norm topology. (Note that any nonmed space may be embedded in its completion, with the given norm extended by continuity; the completion is ipso facto a Banach space.)

Henceforth we shall assume that our topological vector spaces are Ti (and hence Hausdorff, by Proposition 1-3). This is equivalent to the assertion that (0) is a closed subset. For a topological vector space V over k, we distinguish Aut(V), the group of vector space automorphisms V-*V from Aut,,,P(V), the group of topological automorphisms V-4 V (i.e., continuous vector space automorphisms with continuous inverse).

Recall that a subset S of a real or complex vector space is called convex if for every x, yeS, each point of the form tx+(1-t)y, Osts 1, also lies in S. A real or complex topological space is called locally convex if there is a base for the topology consisting of convex sets. Thus, for example, the topological vector spaces R" and C" are both locally convex. DEFINITION. Let G be a locally compact topological group and let V be a locally

convex topological vector space over C. Then an abstract representation of G is merely a homomorphism p:G-+Aut(V). We call p a topological representation (or simply a representation, without qualifier) if it satisfies the additional condition that the map

48

2. Some Representation Theory

GxV--> V (g,x) H Pg(x)

is continuous with respect to the product topology on G x V. [Note that forge G we usually write pg for p(g).]

It follows at once from the definition that for a topological representation p, the image of G under pin fact lies in Auto( V).

2-1 PROPOSITION. An abstract representation p: G-*Aut( V) is moreover a topological representation of G if and only if it satisfies the following two conditions: (i) For every compact subset K of G, the collection of functions p(K) is equicontinuous on V.

(ii) For every xe V, the map g N pg (x) is continuous from G to V. PROOF. =) Certainly a topological representation satisfies (ii), so we need only argue for (i). Let U be a neighborhood of 0 in V. By continuity, for each gEG, there exists a neighborhood H. of g in G and a neighborhood W. of 0 in V such that ph(x)e U for all h eHg and xe Wg. Since K is compact, there is a finite subcollection H,,...,H,, of the H. that cover K. Let W,,..., W. be the corresponding neighborhoods of 0 in V, and set n

W=nW.

.

Then for all gEK and xEW, by construction pg(x)eU, and therefore the collection p(K) is equicontinuous, as claimed.

G) Let (g,x) lie in GxV. Since V is locally convex, it suffices to show that for any convex neighborhood U of 0 in V, there exist neighborhoods H of g in G and W of 0 in V such that for all h EH, p,,(x+W)g pg(x)+ U.

Assume that Kr_ G is a compact neighborhood of g. By condition (i), there exists a neighborhood W of 0 in V such that ph(w)e U/2 for all heK and we W. By condition (ii), there exists a neighborhood H of g contained in K such that for all heH, likewise ph(x)-pg(x)e U/2. Now for arbitrary heG and we V, we have that

Ph(x+w)-pg(x)=Ph(w)+(A(x)-Pg(x))

2.1. Representations of Locally Compact Groups

49

Thus, in particular, if h eH and WE W, then by construction the indicated differ-

ence lies in U/2+U/2. But of course U/2+U/2=U, because U is convex, and this completes the proof.

Note that the set of all mapping from V-+ V is the direct product of topological spaces F1 V V

and thus acquires the product topology, which in this case amounts to the topology of pointwise convergence. The subset Aut(V) in turn acquires the subspace topology, and viewed thus, condition (ii) above implies that the representation p: G--*Aut( V) is a continuous mapping. Therefore, given any compact subspace K of G, p(K) is compact. Consequently, if V is a Banach space, the Banach-Steinhaus theorem implies that p(K) is equicontinuous. Thus we have proved the following corollary:

2-2 COROLLARY. Suppose that V is a Banach space. Then an abstract repre-

sentation p: G-4Aut(V) is moreover a topological representation if and only if for every xe V the map g H pg (x) is continuous from G to V.

REMARK. The corollary holds more generally if V is a barreled space. See Bourbaki, Topological Vector Spaces, Chapter III, § 4.2.

Let p:G-+Vbe an abstract representation of G. A subspace W of V is called p(G)-invariant (or simply G-invariant, when p is understood from the context) if p5(W) c W for all geG. Equivalently, if we view V as a module over the group algebra C[G], then a p(G)-invariant subspace is exactly a C[G]-submodule. Both the trivial subspace (0) and V itself are p(G)-invariant. The class of representations for which these are the only such invariant subspaces is especially noteworthy. DEFINITION. An abstract representation (p, V) is called algebraically irreducible if it admits no proper, nontrivial p(G)-invariant subspaces. A topological representation (p, V) is called topologically irreducible (or simply irreducible, without qualifier) if it admits no closed, proper, nontrivial p(G)-invariant subspaces.

Algebraic irreducibility of course implies topological irreducibility, but not conversely.

Given a representation (AV) of G, we can vary p by any homeomorphic change of basis to obtain another representation that is essentially the same

50


object. We generalize this notion of equivalence just slightly in the following definition to accommodate the possibility of distinct representation spaces:

DEFINITION. We call two representations (p, V) and (p', V') equivalent and V) if there exists a topological isomorphism T: V-+V'such that write (p,

Tops=ppoT

(2.1)

for all geG; that is, for all geG, the following diagram commutes: V'

V Pe I

V

I Ps

T

V.

One checks easily that Eq. 2.1 amounts to the assertion that T is a topological isomorphism of C[G]-modules. Accordingly, we sometimes call T a Gisomorphism. (More generally, an arbitrary linear transformation from V to r that respects the action of G is called G-linear.)

2.2 Banach Algebras and the Gelfand Transform Let A and B be Banach spaces defined over the same field. Recall that a linear transformation T from A to B is called a bounded operator if there exists a real constant c such that IIT(a)II s cllall

(2.2)

for all aeA. It is well known that a linear transformation T is a bounded operator if and only if T is continuous. Henceforth Hom(A,B) denotes the space of all bounded operators from A to B. If TeHom(A,B), then the smallest c that makes inequality 2.2 true is called the norm of T and denoted II TII. One shows easily that Hom(A,B) is itself a Banach space with respect to this norm. In the special case A=B, we write End(A) for Hom(A,A). [Keep in mind that the morphisms in Hom(A,B) and End(A) are always topological as well as algebraic.] Let A be a complex algebra that also admits the structure of a complex Ba-

nach space. Then A is called a Banach algebra if the norm is also submultiplicative; that is, if IIabII s

(2.3)

for all a,baA. Throughout, we assume that our Banach algebras are unital; this is to say that A contains a multiplicative identity 1 e. As usual, the group of

2.2. Banach Algebras and the Gelfand Transform

51

units of A will be denoted A". We can always renorm A without disturbing its topology to arrange that II 'All= 1, and henceforth we do so. (See Exercises 2 and 3 below.)

If A is a Banach algebra, each a(-=A acts on A by left multiplication. Let us denote this map pa. Then according to the inequality 2.3, for all beA, we have that ilpa(b)II=llabllsllall-llbll, whence lIPallsllall, the former norm being computed of course in End(A). Since we assume that IIIA11=11 also IlaII=IIPa(1A)IISWalland thus the norm of a as an element of A agrees with its norm as an element of End(A). Again let a eA and assume now that llall < 1. Then one shows easily that the

series E'o aj converges (see Exercise 4 below), whence we observe that (1-a) lies in A. with

Ea

(2.4)

.

j=o

We shall need this observation for the following result.

2-3 PRoposiTloN. Let A be a Banach algebra as above. Then A" is an open subset of A. Moreover, the mapping

A" -+ A" a -3a-1 is a homeomorphism.

and suppose that for b eA we have that Ila-bll < Ila-' II-' . Then it follows that Ilar'(a-b)ll < 1, whence by the preceding observation we find that

PROOF. Let

the difference 1-a-'(a-b) lies in A". But then also b=a(l-a-'(a-b))eA", showing that A" is open. The second statement follows at once, since the map a I-- a-' is continuous on A" and is its own inverse. With these preliminaries in hand, we now come to one of the principal definitions of this section, essentially a generalization of the notion of an eigenvalue familiar from linear algebra.

DEFINITION. Let A be a complex Banach algebra and let aeA. Then the spectrum of A, denoted sp(a), is the subset of C defined as follows:

sp(a)={A

-a oA"}

.

We shall see below that the spectrum of an element aeA is never empty. Hence we may define r(a), the spectral radius of a, by r(a) = sup(IA1: Aesp(a))

.

52


[For the moment, we can take the spectral radius to be 0 if sp(a) is empty.) The resolvent set of a is the complement of sp(a) in C. By construction, if A lies in the resolvent set, then (A lA-a)-' exists in A.

2-4 PROPOSITION. Let A be a complex .anach algebra as above, and let p(x) be a polynomial with complex coefficients. Then for all aeA, if Ar= sp(a), then p(A) E sp(p(a)).

PROOF. Suppose that p(x) = Eo a,x'. Then we may compute that

iwhere b is some element of the algebra A for which we need no explicit calculation, but only the modest observation that b commutes with a. The point is

this: if the left-hand side of the preceding equation has inverse c, then 24-a has inverse bc, a contradiction, since A is assumed to lie in the spectrum of a. REMARK. This result generalizes to convergent power series over C. (See Exercise 5 below.) 2-5 LEMMA. Let aeA. Then r(a) 5 inf II

PROOF. We first show that sp(a) lies in the closed disk around zero of radius Ilall. Note that in general for nonzero A we have (A - IA - a) = A(l4 -2 a). Thus if IAl>Ilall, Eq. 2.4 applies to show that (A - 1A -a) is invertible. Now let AE sp(a). Then by the previous proposition, A"esp(a") for all n;->O, and therefore, by the first part of the argument, lAI"Slla"ll. Taking nth roots yields the stated inequality.

The following theorem is the first major result about the spectrum of an element. The proof requires three substantial, but familiar, results: Liouville's theorem, the Nahn-Banach theorem, and the Cauchy integral formula. Recall that if A is a complex Banach space, then A *, the dual space, denotes the space of all continuous (equivalently, bounded) linear maps from A to C. 2-6 THEOREM. LetA be a complex unital Banach algebra. Then for every a(-=A sp(a) is nonempty and compact. Moreover, the sequence lla"II"" converges to the spectral radius of a.

PROOF. We first show that the spectrum is at least compact. Consider the continuous mapping


53

C-+A The resolvent of a is simply the inverse image of A" under this map. But then since A" is open, so is the resolvent. Consequently the spectrum of a is closed and, according to the previous result, also bounded. Therefore sp(a) is compact. We next show that sp(a) is nonempty. Fixing an arbitrary VF_A *, define a complex-valued function f on the resolvent set of a by the formula

Note that for fp sufficiently close to zero, we have

f(A-,u) _ q([(2-p).'A -al-' ) = rP([(A..IA-a)(14-

rP(Y u"(A, l4 -a)-"-' ) -o n=0

p"q,((A-1,-a)

n 1)

(The last step follows from the linearity and continuity off) Thus f has a valid power series expansion at every point of its domain and is accordingly holomorphic. Moreover, if 121>11all, we have

f(.)=q((A-14-a)') ='P(A-' (l,, - A-'a)-' )

= 97(Y A-"-'a) n=0

"=0

and we can therefore bound f as follows: YIAI-"-' IIq, -0

II(v 11

IAI-Ilall

(2.5)

54


Now assume that the spectrum of a is empty, whence a is nonzero. Then f is entire, bounded on the closed disk I1IS211aII by general principles, and bounded elsewhere by the quotient IIrvII1IIaII according to the previous inequality. By Liouville's theorem, f must be constant, and since clearly lim f(2) -- 0 as I2 I --> oo, this constant must be 0. Since this holds for arbitrary rpeA *, it follows from

the Hahn-Banach theorem that (A IA - a)-' is 0, which is impossible. Hence sp(a) is nonempty.

Finally, it remains to establish that the spectral radius of a is as stated, and in establishing this, we may certainly assume that an is nonzero for all nEN. First we claim that the power series expansion for fgiven in Eq. 2.5, which was established for I.%l>IlaJJ, in fact holds with uniform convergence for JAIzr, for all r greater than the spectral radius of a. To see this, consider the auxiliary function

g(A)

_{f(A-) forA4, 0 otherwise.

0

Since as we have seen, f is holomorphic for JAlzr> r(a), the power series representation

g(2) = I An+Ip(a") n=0

extends to the entire closed disk IAISr-l. Moreover, the Cauchy integral formula tells us that for JAI Sr-' the remainder g,,, l after n terms is given by A.al

2'ri,

g(` ) +l

ds

where the integral is taken over the circle C of radius strictly between r-l and r(a)-l. It follows easily from this that this remainder is bounded independently of A. The upshot is that since g is represented by the uniformly convergent power series given above for I2I5r-1, f is correspondingly represented by the power series representation of Eq. 2.5 with uniform convergence for I2Izr, as claimed.

Next let A=re'B where r>r(a). We may then integrate the series for A-"f(2) with respect to Oas follows: 2a

Jrn+lei(n+l)8f(re'B)d© 0

_ 2x

= I f rn-me,(n m)eco(am)dOn m=00

= 2,rp(a")


55

Moreover, this value is clearly bounded by 2,rr"''M(r)II q ll, where

M(r)=sup11 re'a'lA-all e

Thus

I p(an)l!5 r"'' M(r)II rp I

I

Appealing again to the Hahn-Banach theorem, we see that the linear mapping ya" Hylla"ll (yEC), which is obviously of norm 1, extends

for all

from the one-dimensional subspace spanned by a"(*0) to an element SoeA* of lesser or equal norm. In this special case, the previous inequality reduces to

lla"ll 5 r' M(r)

.

Since this holds for all r> r(a), taking nth roots and limits we find that lim sup Ila"ll't" 5 r(a)

.

This inequality together with Lemma 2-5 shows that the sequence Ila"ll'" is indeed convergent to the spectral radius of a. 2-7 COROLLARY. (Gelfand-Mazur) IfA is a division ring, then A-C.

PROOF. Given aeA, there exists 2.Esp(a), so that A IA- a is not invertible. But if A is a division ring, then A. lA-a = O, whence every element of A takes the form for some complex A. Then evidently, A-C.

Quotient Algebras In preparation for the discussion of the Gelfand transform, we make some brief remarks on the quotient of a Banach algebra A by a (two-sided) ideal J, which in particular is a linear subspace of A. Recall that as an algebra, AU consists of the cosets a+J. We say that a represents its associated coset, and addition and multiplication of cosets are defined by the addition and multiplication of associated representatives. We define a seminorm on A /J by the formula lha+Jll = inf{Ila--xll : xEJ}

.

(2.6)

It is easy to see that this is well-defined and lacks being a norm only insofar as it is possible that 11a+Jll=O without it being the case that a represents the zero element of the quotient.

56


2-8 PROPOSITION. Assume that J is closed in A. Then Eq. 2.6 defines a norm on A U. and A /J is likewise a Banach algebra with respect to this norm.

PROOF. In light of the preceding remarks, it suffices to show that the seminorm on the quotient is submultiplicative and yields zero only on the zero element of the quotient space. We consider first the latter point. If IIa+JII=O, there must exist a sequence of points in J converging to a. But since J is assumed closed,

this means that aeJ, whence a+J=J, as required. It remains to show that the seminorm on A /J is submultiplicative; that is, IIab+JII

First note that since J is a linear subspace, IIa+JII can equally well be defined as inf{Ila+xll : xEJ}. Accordingly, Ila+xII

ynfllb+yll

z x.inf Ilab+xb+ay+xyll z XEJ infllab+xll =IIab+JII The first inequality in the calculation is justified by the submultiplicative nature of the norm on A; the second is justified because the sum xb+ay+xy clearly lies in the ideal J, provided that x and y do. This completes the proof. REMARK. Note that if J is an ideal of the Banach algebra A, then in particular, J

is a subgroup of a topological group, and we may infer from Proposition 1-1 that the closure of J is likewise a subgroup of A. Moreover, since the norm is submultiplicative, if {x.} is a convergent sequence in J, then so are the sequences {ax.} and {xla} for all aeA. It follows that the closure of J is likewise an ideal of A.

The Gelfand Transform In this subsection we specialize to commutative complex Banach algebras (always assumed to be unital). If A is such an algebra, a character of A is simply a nontrivial (hence surjective and unital) homomorphism of complex algebras from A to C. The set of all characters of A is denoted A. 2-9 PROPOSITION. Let A be as above. Then the following statements hold:

(i) Every maximal ideal ofA is closed.


57

(ii) The mapping y H Ker y constitutes a h yective correspondence between A and the set of maximal ideals of A.

(iii)Every element ofA is continuous (iv) For every aeA, sp(a) = { y(a) : yEA ).

PROOF. (i) Let M be a (two-sided) maximal ideal of A; that is, M is a proper ideal of A and there exist no ideals properly between Al and A. By the remark

above, M, the closure of M, is likewise an ideal of A, and so to show that M = M it suffices to show that Al is a proper ideal; that is, that Al excludes all units. But since A' is open by Proposition 2-3, any unit in Al must be the limit of units already included in M, contradicting the assumption that MxA. Hence the maximal ideal M is closed, as claimed.

(ii) Since every character y is surjective, the quotient A /Ker y is a field. Hence Ker y is maximal, and the given mapping is at least well-defined. Let M be the closed ideal Ker y. Then we have the following commutative diagram:

r A --C AIM

Here p denotes the canonical projection onto the quotient (a continuous homomorphism of Banach algebras), and y is the unique induced map on the quotient, which is at least an isomorphism of complex algebras. Every element of

AIM takes the form z lA+M for some zeC, and in fact the induced isomorphism is precisely

y is, moreover, continuous: for open UcC, Y-'(U) which is evidently open in AIM. Conversely, if M is any maximal ideal of A, then AIM is not only a Banach

algebra but also a field, which by Corollary 2-7 is isomorphic to C. Call this isomorphism y,,,. Then the diagram above defines a character y,, = YM °P, and it is straightforward to check that for all characters y, YKerr=Y and for all maximal ideals M,

58


Ker y,,, = M . This establishes the claim.

(iii) The continuity of an arbitrary character y follows at once from its factorization above into two continuous maps.

(iv) Let aEA. Then Aesp(a) if and only if (A 1A-a) is not a unit of A, and hence (by Zorn's Lemma) if and only if (A- lA-a) is contained in some maximal ideal lies in the kernel of some M. But by part (ii) this occurs if and only if character y, which is to say, if and only if y(a)=A for some y. We next introduce a topology on A, the space of characters on A, by duality. As a preliminary, note that for each aeA, we have an associated map from A* to C defined by q H rp(a) ; this is simply evaluation at a. Recall that one then defines the weak-star topology on A * (abbreviated to the w*- topology on A *) to

be the weakest topology on A with respect to which all such evaluation maps are continuous. Under this topology A* is a locally convex topological vector space and, in particular, Hausdorff. (See Appendix A; especially sections A.2 and A.3.) Moreover, convergence in the w*-topology amounts precisely to pointwise convergence. Part (iii) of the previous result shows that in fact A lies in A *. The subspace topology on ,4 induced by the w*-topology on A * is called the Gelfand topology on A.

2-10 LEMMA. The space A of characters on A lies in the unit ball of the dual space A*. Moreover, A is both Hausdorff and compact with respect to the Gelfand topology.

PROOF. For each aEA and yeA, we see from Proposition 2-5 and Proposition 29, part (iv), that y(a) 5 r(a) 0, the set {xeX : Lf(x)) - e) is compact. If X'=Xv{ w} is the Alexandroff one-point compactification of X, then it is easily verified that fEW0(X) if and only if f extends to a continuous complex-valued function f on '(X') such that f(w)=0. 2-13 COROLLARY. Let X be a locally compact Hausdorfspace and let A be a self-adjoint subalgebra of 'o(X) that separates points with the additional property that for every xeX there exists an feA such that f(x)*0. Then A is uniformly dense in 8'0(X) with respect to the sup norm. PROOF. Again let X' denote the one-point compactification of X. (Note that this makes sense even if X is already compact, in which case we have simply adjoined an isolated point.) Identify A with a subalgebra of 8'(X') by extending each element to a function that vanishes at co, and let A' be the subalgebra of W(X') generated by A and the complex constant functions. Then A' is evidently self-adjoint and unital. Moreover, A' separates points: since A already separates

2.3. The Spectral Theorems

61

points in X, we need only observe that by hypothesis, for every xeX there is a function feA that does not vanish at x, while by construction its extension to X' does vanish at w The previous result now applies to show that A' is uniformly dense in W(X'). Thus for each ge'o(X) [tacitly identified with an element of ' '(X')] and for each positive a there exists an fEA [again identified with an element of '(X')] and a a .EC such that I g(x) - f (x) +Al < e/2

for all xoX'. Since both f and g vanish at o it follows that JAI < e/2, and therefore f and g differ on X by less than e, as required.

Bounded Operators on Hilbert Spaces In this subsection we specialize our analysis to the Banach algebra of bounded operators on a Hilbert space. Actually, only a few formal aspects of such an algebra will be needed, and these we highlight below. First recall that a positive definite Hermitian form on a complex vector space H is a mapping

HxH-* C (v,w) H (vlw) that satisfies the following properties: (i)

(u l u) E R, (u a H), with equality if and only if u = 0

(ii)

(ulv)=(vju) (u,v eH)

(iii)

(A.u + ftv l w) = 2(u l w) + p(v l w)

(u, v, w H; ,Q, p E C)

Note that (ii) and (iii) imply also:

(iii)' (uI2v+,uw)=A.(uIv)+fs(uIw)

(u,v,weH;A,pcC)

That is, the form (I) is positive definite, conjugate symmetric, linear in the first variable, and conjugate linear in the second. A complex vector space H together with a positive definite Hermitian form is called a pre-Hilbert space. One shows easily that (I) defines a norm on H as follows: Ilvli =

(vlv)

.

62


If H is moreover complete with respect to the associated metric, then H is called a Hilbert space. In particular, H is a complex Banach space, and therefore a topological vector space with respect to the topology induced by the norm. Assume for the remainder of this discussion that H is a Hilbert space, and in accordance with previous usage let End(H) denote the space of bounded linear maps from H to itself. End(H) is thus a Banach algebra with respect to addition

and composition of functions, and it acquires some significant new structure from H. In particular, it is well known (see Exercises 9 and 10 below) that for called the adjoint every TeEnd(H) there exists a unique element of T, such that

(Txly)=(xIT*y) for all x,yEff. Moreover, the adjoint has the following elementary properties: (i) For all TeEnd(H), T**=T; that is, the adjoint operator has period two.

(ii) For all T,,T2EEnd(H) and 2,,.12EC, (d7, +A2T2)*=;i T,* +A.2T2*; that is, the adjoint operator is conjugate linear. (iii) For all T,, T2EEnd(H), (T, T2 )* = T2 * T, *; that is, the adjoint operator is antimultiplicative. (iv) For all TEEnd(H), II T11 = 11 T* II; that is, the adjoint operator is an isometry; in particular, the adjoint operator is continuous. (v) For all TeEnd(H), I17' T* II = II TII2.

The usual arguments from linear algebra suffice to establish properties (i)(iii). To establish (iv) and (v), note that for all T, IIT(x)112= (T(x)IT(x))= (7'*T(x)Ix)!9 This shows that IITII25IIT*TII. But also IIT*TIISIITII'II7'*II, so we have the chain of inequalities IITII25IIT*

TII' denote the invariant scalar products on V and V', respectively.

82


21. Let (,r, V) be an irreducible unitary representation of a compact group G with Haar measure dg. Verify the following identity for all vj,v2,v3,v4E V:

.1(2r(g)v1Iv2)(;r(g)v3Iv4) dg

dim(V)(vllv2)(v3Iv4)

G

22. Let G be a locally compact group with Haar measure dg. Define L2(G) to be the Hilbert space of square-integrable functions on G; that is, L2(G) consists of the measurable functions f. G-3. C such that

JIr(g)I2 dg jfrpds

It follows at once from this that IIq?1I.51 and that

55q (s't)f(s)dsf(t)dt--

jjrp(s-'t)f(s)dsf(t)dt

whence rp is also of positive type and therefore continuous by the previous cor-

ollary. More precisely, to represents the equivalence class of a continuous function in L-(G)]. Thus as a subset of L'(X)*, 3'(G) corresponds to a closed

3.2. Functions of Positive Type

99

subset of the unit ball, which is therefore compact under the weak-star topology by Alaoglu's theorem. (ii) This is a special case of the Krein-Milman theorem.

(iii) The only point to check is that a nonzero extreme point q e.,P(G) satisfies rp(e)=1. But if ip(e),, = A(s)r(e) = A(s)

whence to is likewise a character, as claimed.

3.3 The Fourier Inversion Formula The principal technical tool for establishing the Pontryagin duality theorem in the following section is the Fourier inversion formula. In this section we review the Fourier transform and prove this fundamental result. Throughout, G denotes a locally compact abelian group with bi-invariant Haar measure dr and continuous complex character group G. DEFINT11oN. Let feL'(G). Then we define f:G -+ C, the Fourier transform of f, by the formula

I(X) = f f(y)x(y)dy for %E G.

Note that this formula makes sense, since for all yeG, X(y) has norm 1. Hence if f is integrable, so is the product appearing in the integrabd. Moreover, one verifies at once that l f(X)15 II f14 for f e L' (G), X e 6. REMARK. In the special case that G=R, the topological group of real numbers with respect to addition, we can identify each teR with the character

sHe

3.3. The Fourier Inversion Formula

103

In this case the formula above reduces to

f(t) = j.If

(s)e-r:ids

R

which is of course the ordinary Fourier transform of a function defined on R. The point is that despite appearances, this should in fact be regarded as a function on A. Let V(G) denote the complex span of the continuous functions of positive type on G, and define

V'(G) = V(G)rL'(G)

.

We can now state the principal result of this section. (See Exercises 13 and 14 below for direct proofs of this theorem and the duality theorem for G finite.) 3-9 THEOREM. (The Fourier Inversion Formula) There exists a Haar measure dZ on G such that for all feV'(G),

.f(Y)= J.f(X)X(Y)dX Moreover, the Fourier transform f H f identifies V'(G) with V'((§). The measure dX of the theorem is called the dual measure of dx, the given Haar measure on G. To prove its existence, we must begin with some elementary properties of convolution. 3-10 PRoPOStTION. Let f and g be complex-valued Bore! functions on the locally compact abelian group G. Then the following statements hold. (i)

If the convolution f*g(x) exists for some xeG, then so does g*f(x), and in fact g*f(x) = f*g(x).

(ii) If f'geL'(G), then f*g(x) exists for almost all xeG; moreover,

f*geL'(G) and Ilf*gll, 5 VIII, II9II1

-

(iii) Iffg, heL'(G), then (f*g)*h=f*(g*h). Thus, in particular, convolution is both associative and commutative on L' (G ).

104

3. Duality for Locally Compact Abelian Groups

PROOF. (1) This follows by direct application of the translation-invariance of the Haar measure on G. We replace y by yx in the integrand that defines convolution to obtain

f *g(x) = f g(y 'x)f(Y)dy = f g(Y ' )f (Yx)dy

=g*f(x) Note that the last step is justified by the elementary observation that for locally compact abelian groups, the Haar measure of a Borel subset E of G is equal to that of E-1. (See Exercise 7 below.) (ii) First consider the homeomorphism a from GxG to itself defined by a(x,y) = (Yx,Y)

Observe that the inverse map sends (x,y) to (y'x,y). Next consider an open subset US C. Then a(f-'(U)xG) is clearly a Borel subset of GxG, and by construction, (x,y)Ea(f-'(U)xG) if and only ify'xe f-'(U). This shows that the mapping

(X, Y) H f(Y-'x) is a Borel function on GxG and hence so is

(X, Y) N f(y 'x)g(Y) since the product of Borel functions is again a Borel function. (Here we may view g as a function on GxG in the obvious way.) Since both f and g are L'functions, we have If I f(y-'x)I d I g(Y)I dy
h

X H vX is a bijection.

Note that the proposition subsumes the assertion that the Fourier transform of the convolution f*g is the complex product of Fourier transforms f g .

106


Pttoop. Clearly each vx is linear on L'(G), and not identically zero, since each character x of G takes values of norm 1. We check with a routine calculation that each such map is multiplicative:

vr(f*g)= f f*g(y)x(y)dy

f(z 'y)g(z)d=x(y) dy =j

jf(z-'y)x(y)dyg(z)dz

_ f jf(y)x(zy)dyg(z)dz = f f (y)x(y) dy j g(z)x(

dz

= f(x)8(x) We show next that every nonzero character of B is of the form vx for some group character X. Let lv:B--i-C be a nontrivial algebra homomorphism. By Gelfand theory (Lemma 2-10) we know that yr is a functional on L'(G) of norm bounded by 1. Hence by the duality of L' and L`° there exists some gML`°(G) having identical norm such that

w(f) = f f(x)q,(x)dx a

for all feL'(G). Recall that for any yeG and function f defined on G, Lyf is defined by Ly f(x) f(y-'x). Now compute:

j V(f)g(y)w(y) dy = w(f)w(g)

= y(f *g) = J I f(y 'x)g(y)dyvo(x)dx

= j5L,.f(x)p(x)dxg(y)dy = j w(Lyf)g(y)dy Thus we have that

w(f)w(y) = w(L,f)

(3.4)

for almost all yeG. One shows readily that the expression on the right is continuous in y-the elements of WA(G) are dense in L'(G) and left and right uniformly continuous-whence we may assume that w is likewise continuous; here


107

we need that yi is not zero. Now applying the previous equation three times, we obtain

w(f)c(xy) =

w(L=L,.I) = w(L,.f)c(x) = w(f) v(X)go(y)

Again since w is nonzero, c is multiplicative on G. Thus in particular,

4,(y-') = ta(y)-i whence Ic,(y)(=1 for all ye G, because V, has L°°-norm bounded by 1. This shows that 9 is indeed a character of G and that yv = iV .

Finally, given two group characters X and z', if vx(f) = ve (f) for all functions feL'(G), then by duality, X and z' must agree almost everywhere in G. But since both are continuous by definition, it follows that X=X', as required. 7

The Ring of Fourier Transforms and the Transform Topology Consider now the space A for, more explicitly, A(G), should we wish to emphasize the underlying locally compact abelian group G[ defined by

A={f:f EL'(G)}

.

Thus A consists of the Fourier transforms of functions feL'(G) and incidentally defines a weak topology on G, the space of complex characters of G; this is the weakest topology such that each f eA is continuous. We shall call this the transform topology on 6. Since the Fourier transform of f *g is the complex product of functions f g , it follows that A in fact constitutes a ring of continuous functions on G with respect to the transform topology. Now, according to the previous proposition, each element f r =A may be regarded as the Gelfand transform of f insofar as we identify G with the space of characters on L'(G) via the mapping Z" vx. More precisely, we have by construction that

f'(vx) = i5(f) = .?(Z) where, strictly speaking, on the left f denotes the Gelfand transform operating on the space of characters of L'(G) in the sense of Chapter 2, and on the right f denotes the Fourier transform operating on G. These considerations lead at once to the following proposition.

108


3-12 PROPOSITION. Let G have the transform topology induced by A . Then the

ring A is a separating, self-adjoint, dense subalgebra ofTo(G). PROOF. Let us first consider h = Horn c(B, C)* . According to Lemma 2-10, if L'(G) is unital, then h is a (weakly) compact subset of the dual space B*, and given fEL'(G), its Gelfand transform f lies in '(B) . Otherwise, B perhaps is not closed, because the weak limit of nontrivial algebra homomorphisms may in fact be trivial. Nonetheless, in either case B' = B u(O) is closed. Thus for

each feL'(G), we have that f , when extended by zero to h', lies in '(B'), and therefore f e W o (B) .

Now identify W (d) with T(B) according to the topological isomorphism induced by Z H vX. Then by Gelfand theory (Theorem 2-11 and Exercise 8 of Chapter 2) it follows that A is at least contained in Wo(G) and separates points. Thus it only remains to show that A is self-adjoint, since its density in '0(G) is then a consequence of the Stone-Weierstrass theorem (see Proposition 2-13 and Exercise 9 below). Let feL'(G). Then for all characters X on G, we have '

j.r(y ')x(y)dy = 5f(y)x(y-' )dy =

f f(y)x(y)dy

= AX) showing that A is indeed closed under complex conjugation, as required.

This application of Gelfand theory becomes even more compelling in consideration of the following theorem: 3-13 THEOREM. Let G and G be as above, and let K denote a compact subset of G, and Van open neighborhood of 1 in S'. Then the following statements hold.-

(i)

Each of the sets W(K, V) as defined in Section 3.1 is an open subset of d in the transform topology.

(ii) The system (W(K, V)) in fact constitutes a neighborhood base for

the trivial character with respect to the transform topology of G.

(iii) The compact-open topology and the transform topology on G are identical.

Note that (ii) immediately implies (iii), since by construction {W(K,V)} is also a neighborhood base for the trivial character with respect to the compactopen topology. The proof will be straightforward, given the preliminary lemma that follows.


109

3-14 LEMMA. Let G x G have the product topology defined by the topology given on G and the corresponding transform topology on G. Then (i)

For everyfeL'(G), the map

GxG-+C (Y,x) H (L,.f)^ (x) is continuous. [Here (L, ,f )A denotes the Fourier transform of the left translation of f by y.] (ii)

The map

GxC; -iC (Y, X) I-) %(Y)

is likewise continuous.

PuooF. (i) Let (y0,

be any fixed point in the domain of the given map. Then, according to Exercise 8 below, for every a>0, there exists a neighborhood U ofy0 such that

IILn-L,,II<e for all ye U. Moreover, by construction of the transform topology there exists a neighborhood V of Xo such that

I(L,,,f)^(,) - (L,bf)^(Xo)I < e

for all xeV Now since for any L'-function g and character x, I8(x)ISIISII1, it follows in particular that

I(L"f)^W - (L f)^(x)I s Iw,J- L,b11I1

.

Therefore,

I(L,.f)^cz)-cL,

f)^(xo)ls2e

whenever (y,x)e UxV, and this clearly establishes the asserted continuity.

(ii) Note that Eq. 3.4 is equivalent, in the special case q' = X, to the equation

I(x)x(Y) = (L,f)^ (x)

110


Exercise 9 shows that for every ,Z there is an L'-function whose Fourier transform does not vanish at x and hence does not vanish in a neighborhood of x under the transform topology. Thus this last equation implies that the function

(y. x) H x(y) on the product space is, according to part (i), the quotient of two continuous functions. Therefore both this function and its conjugate are likewise continuous. This completes the proof. PROOF OF THEOREM 3-13. According to our preliminary remarks, we need only prove parts (i) and (ii). Moreover, it clearly suffices to deal with subsets W(K, V)

of the form W(K,N(s)), where the neighborhoods of the identity N(e)cS', e>0, are defined as in Section 3.1.

(i) Let K be a compact subset of G, and let a>0 be given; choose and fix %oEW(K,N(s)). Then in consequence of the preceding lemma, for every yoeK there exist open neighborhoods U of yo in G and V of xo in G (with respect to the transform topology) such that x(y)EN(e) for all xe V and ye U. The compact subset K is covered by finitely many open sets U,,..., U,. with correspond-

ing character sets V,,..., V, Clearly the intersection of the Vi is an open neighborhood of Xo contained in W(K,N(e)), and therefore W(K,N(e)) is open in G, as claimed. (ii) Let V be an open neighborhood of the trivial character, here denoted 1. We must show that V contains a subset of the form W(K,N(e)) for some compact subset K of G and some positive a But by definition of the transform topology (consider its subbasel) we know that for some s,>0 there must indeed exist a finite family of functions f ,..., f EL'(G) such that

n{x:If,(x)-f;(1)I< }sV

.

Since ''(G) is dense in L'(G), we may further assume, at the cost of decreas. ing e,, that each of the j has compact support K,. Let K denote the (necessarily compact) union of the Kj and choose a positive e subject to the inequality e
1 K

and hence the corresponding integral over the complement of k is less than e. Now consider the identity

g(y) = f S(X)X(y)dd+ f 8(X)X(y)dX x

x°

given by the Fourier inversion formula. As. V shrinks to a sufficiently small neighborhood of 1 in S`, the first integral above eventually lies within s of unity for all ye W0(K,V), while the second is unconditionally bounded in absolute value by e. Hence g must be bounded from below by 1-2son W0(K, V). But by construction, U contains the support of g, and therefore U contains W0 (K, V), thus completing the proof.

3-23 COROLLARY. The mapping a defined above is bicontinuous; thus a is a homeomorphism onto its image. PROOF. By construction we have the identity

a(W0(K,V)) = W(K,V)n a(G) which in light of the lemma and the proposition shows that a is bicontinuous at the identity element of G. Since a is clearly a group isomorphism onto its image, the result holds everywhere in G by translation.

3.4. Pontryagin Duality

121

Recalling one of the fundamental facts of topological groups, this first corollary nets us a second: 3-24 COROLLARY. The image of a is closed in G.

PROOF. By general topology, a locally compact and dense subset of a Hausdorff

space must be open. Now a(G) is locally compact, being the homeomorphic image of the locally compact group G, and, of course, is dense in its closure in the double dual. Accordingly, a(G) is an open subgroup of its closure. But since every open subgroup of a topological group is also closed, a(G) is in fact identical to its closure, as required.

Given these two corollaries, the proof of Pontryagin's theorem reduces to showing that a(G) is dense in the double dual of G. This requires a final bit of delicate analysis.

The Plancherel Theorem Let shows that

for xeG. An easy calculation and as usual, definef(x)=f(x) 1W

I(x)=I(x) Set g = f * f ; then certainly g is integrable and moreover, according to Exercise 5 below, of positive type. ff f lies also in L2(G), the Fourier inversion formula yields the following key observation:

f If(x)I2dx=g(l) = f s(x)dx = f I I(x)I2 dx This shows that the Fourier transform induces a map

L1(G)n L2(G) -- L2(G)

fHI which is an isometry onto its image.

122


Recall that A = A(G) denotes the ring of Fourier transforms of functions in L'(G). Let A, denote the subset of A arising from the isometry above. Note that A, is stable under multiplication by elements of a(G):

[a(yo)-f](X)=X(yo)jf(y). (y)dy = J.f(y)X(yo'y)dy

= Jf(yoy)x(y)dy _ (L, f)^(X) -

The following result is the key to our current discussion.

3-25 LEMMA. A, is a dense subspace of the Hilbert space L2(G).

Granting this, since also L'(G)nL2(G) is dense in L2(G)-the intersection contains ',(G)-the isometry defined by the restricted Fourier transform may be extended by continuity to an isometric isomorphism L2(G) -+ L2(G)

fHf

.

Note that we continue to use the circumflex notation for this extended version of the Fourier transform, called the Plancherel transform. To summarize, relative to the preceding lemma, we have established the following: 3-26 THEOREM. (Plancherel) Let G be a locally compact abelian group. Then

the extended Fourier transform defines an isometry of Hilbert spaces

from L2(G) onto L 2(d).

El

PROOF OF LEMMA. In view of the self-duality of Hilbert spaces and the Hahn-

Banach theorem, it suffices to show that zero is the only element of L2(G) orthogonal to every element of A, . Assume that ge L2(G) is orthogonal to every element in A, Since A, is stable under multiplication by elements of a(G) for all and yEG, we have that .

Jg(X)f(X)X(y)dX=O

.

3.4. Pontryagin Duality

123

This says that the Fourier transform of the measure g(X)f(X)dd is zero, and hence by a slight extension of Proposition 3-15 so is the product gf almost everywhere. But note that for a character x we have (x f)^ = L' f . Thus given any nonzero continuous element of A,, we can produce an element of 4, that does not vanish in some neighborhood of Z. Hence if the product gf is zero almost everywhere, it must be that g is zero in L2(G), as required. 3-27 COROLLARY. (Parseval's Identity) For all fgeL2(G), we have

5f(x)g(x)dx = 5f(X)8(X)dX PROOF. By elementary linear algebra, a linear isometry is necessarily unitary.

3-28 COROLLARY. Let f and g lie in L2(G), and let h lie in L'(G). Then if

h=fg, we have h = f *g. PROOF. Suppose that h factors as given. Let Xo be a character. We compute as follows, appealing to Parseval's identity to justify the transition from the second to the third line: h(Xo) = Jf(y)g(y)X0(y)dy

= Jf(y)g(y)xo(y)dy =

Jf(X)g(x'Xo)dX

I*g(xo) This completes the proof. 3-29 COROLLARY. The ring A of Fourier transforms of L L functions on G consists precisely of convolutions of junctions in L2(G).

PROOF. If heL)(G), then h factors as fg for functions f,geL2(G). For instance, h=r-Irl where rnL2(G) is defined by

r(x) _

h(x) / I h(x)I112

0

if h(x) # 0 otherwise.

124


Hence h = f +g, and therefore every element of A is of the required form. Conversely, by Plancherel's theorem every convolution of functions in L2(G) takes the form f +g for some f and g as above, and hence is the transform of the L'-function f-g. Accordingly, such products lie in A , as required.

This brings us to the final technical prerequisite for the proof of Pontryagin's theorem.

3-30 PRoposmoN. Let U be a nonempty open subset of G. Then there exists a nonzero function f eA with support contained in U. PROOF. Recall from Proposition 1-7, part (iii), that the volume of any nonempty open set relative to a Haar measure is positive. Thus, by inner regularity, there exists a compact subset K of U with positive measure. At every point of xEK we can find an open neighborhood V of the identity and an open neighborhood Ux of x such that UXV, is contained in U. Then since K is compact and G is locally compact, there exists a compact neighborhood V of the identity such that K V is

contained in U. Define f as the convolution of the characteristic functions on K and V, respectively. It follows at once from the previous result that f EA and that f has support contained in KV, and therefore contained in U. Moreover, one calculates at once that the integral of f over G is simply the product of the measures of K and V, and hence positive. Thus f is nonzero on a set of positive measure.

Proof of Pontryagin 's Theorem As we observed above, it remains only to show that a(G) is dense in G. If not, then according to our last proposition, there exists a function in q e L'(G) such

that , is nonzero but nonetheless , vanishes on a(G). Let Xo lie in the double dual. Then by definition,

w(zo) = J c (Z)zo(z ' )dx . But the assumption that

vanishes on a(G) means precisely that 5 q'(x)X(Y-' )dx = 0

Exercises

125

for all yEG. Hence, as in the proof of Plancherel's theorem, OP =0 almost everywhere, and therefore = 0. This contradiction completes the proof.

The only remaining issue in this chapter is to establish the last statement of the Fourier inversion formula, namely that the Fourier transform identifies V'(G) with V'(6). We have already shown that the map f H f is injective. Let F lie in V'(G), and define a function Jon G by the formula

f(y)= JF(X)X(y)dx Identifying G with G, this amounts to

f(y)=F(Y') which places fEV'(G) by Corollary 3-19. [One verifies at once from the defini-

tion that ifyH q'(y) is of positive type, then so is yH q'(y-').] By the Fourier inversion formula,

F(X) = f F(y)X(y) dy

= jf(y-')X(y)dy = jf(y)X(y)dy and this shows that F is the Fourier transform off Hence f H f is also surjective, as required.

Exercises 1.

Let G be a locally compact topological group. We consider functions from G into either the real or complex numbers.

(a) Let f, andf2 be Haar-measurable functions on G. Show that the productfif2 is likewise Haar-measurable on G. (b) Let f be a Haar-measurable function on G. Define F on G xG by

F(g,h) =f(g)f(h) Show that F is Haar-measurable on G xG.

126


(c) Let p be a Haar-measurable function on G. Define w on GxG by

u(g,h) _ q(g'h) Show that v is Haar-measurable on GxG. 2.

Let G be a topological group and let X be a metric space, with xoeX. Suppose that f : G-*X is a continuous function subject to the condition that there exists a compact subset K of G such that if soK, then f(s)=xo. (Thus we generalize the idea of compact support to cases for which the codomain has no algebraic structure.) Use Proposition 1-1 to show that f is uniformly continuous in the following sense: for every a>O, there exists a neighborhood V of the identity in G such that I f(s) f(t)I < e whenever s-'te V.

3.

likewise lies in Wc(G). [Hint: Let feWJG) and let ge'o(G). Show that For continuity, use an easy extension of Exercise 2 to show that if t.-+t in G, then for any positive a eventually

I J(g(s 't)-g(s-'ta))f(s) dsI 0 XEX

Choosing y in R such that 0 0, then k is ultrametric and isomorphic to the field of formal power series in one variable over a finite field (i.e., the quotient field of FQ[[tll for some finite field Fq and indeterminate t). We begin with some preliminary results on topological vector spaces.

Topological Vector Spaces over Nottdiscrete Locally Compact Fields Let V be a topological vector space over a nondiscrete locally compact field k, and let W be a finite-dimensional subspace of V of dimension n. Assume further that W has basis wi,...,w". Consider the map k"

(aj )

V

-+

it'

i-4 I aj wj

Clearly p is a sum of continuous functions, from which one deduces at once that rp is a continuous isomorphism of topological vector spaces.

4-13 PROPOSITION. Given k, V, W, and rp as above, the following assertions hold: (i)

Let U be any open neighborhood of zero in V. Then Wrn U# {0}.

(ii)

The mapping to is a homeomorphism. Consequently W admits precisely one structure as a topological vector space over k.

4.2. The Classification of Locally Compact Fields

141

(iii) W is closed and locally compact.

(iv) If V is itself locally compact, then V is finite-dimensional over k, and modv(a)=modk(a)'imV for all aeV PROOF. (i) Wn U must contain something other than zero, else via V-1 the zero vector would constitute an open subset of k", contradicting the assumption that k is not discrete. (ii) We need only show that rp is an open mapping. Since according to Proposition 4-7, the sets B I _ {aek: modk(a)5 t}

(t>O)

constitute a local base at zero for k, it suffices to show that for all positive t, ip(B$) contains a neighborhood of Oe W. We introduce an auxiliary map w

k"

-+

R"

(ai) H (modk(ai))i:i...

"

which is continuous by Proposition 4-1. Define subsets A of k" and X of R" respectively, by

A = ((ai) e k": sup (modk (ai )) =1)

i

and

X=U{(xi)ER;:x;=l,xxSlforjxi} Clearly neither set contains zero. Note, moreover, that X is closed in R", and therefore A, which is precisely yr1(X), is likewise closed in k". Furthermore, A is a subset of the compact set B and therefore itself compact. Now consider q(A), a compact subset of V, which also does not contain zero. Since scalar multiplication is a continuous map from kx V to V, the inverse image of V-q(A) contains an open neighborhood of (0k,0v). Again, since the sets B1 constitute a local base at zero for k, it follows that there exists an open neighborhood U of zero and an a>0 such that B.Unrp(A)=0. This is to say that ifyek with modky-<e, then yUn o(A)=0. Fix t>O and choose aEk such that 0<modk(a)set. (Such a exist according to Corollary 4-2.) By part (i), the set (WnaU)-{0} is nonempty; suppose that w=Za.w. lies therein. Let h be the index such that modk(ah) is maximal and hence positive. Finally, define the following parameters:

4. The Structure of Arithmetic Fields

142

bj =

(j = 1.... ,n)

z=a. 'w Since (b) lies in A, z lies in rp(A). Since w lies in aU, z lies in yU with y = ah 'a. By definition of Uand ewe must have that modk(y)>e. Therefore, modk(ah) < e-' mod,(a) 5 t

.

This implies that (aj) a B; and that w r= q7(B,). We conclude that

WrnaUcq,(B7) whence q(B,) indeed contains a neighborhood of 0 in W, and therefore 9 is open.

(iii) Clearly W is locally compact by part (ii). Suppose that z lies in the closure of W but not in W itself. Then again by part (ii), we have a homeomorphism of k"" onto <W, z), the subspace generated by W and z, which maps the closed subspace k"x{0} onto W. It follows that W itself is closed in (W,z), whence ze W-a contradiction. Thus W is closed in V, as claimed.

(iv) Assuming for the moment that V is indeed finite-dimensional over k, in light of part (ii) it suffices to prove the formula for mod,(a) for V=k". But by Fubini's theorem, the effect of left multiplication by a on the measure of a measurable subset of k" may be computed iteratively over each of the coordi-

nates, and from this we deduce immediately that mod,(a)=modk(a)", as claimed.

It remains to show that a locally compact topological vector space V is in-

deed of finite dimension over k. Let there be given aek such that 0<modk(a) 0. In particular, setting a1 1 for all j, we obtain the inequality InIs2n. Moreover, i=I

J=1

We may now proceed with the main calculation:

l a+bl"= 1(a+b)"I =I

(7)aib"-i

5 2(n+1)yI(")I IaliIbI" 3 jO 54(n+1)Y,(,)Ialilbl"-i

i=o

=4(n+1)(Ial +IbI)" Taking the nth roots of both sides and then the limit as n-->oo now yields the triangle inequality. Note that if I I is an absolute value, then I1I=1. Indeed, by AV-2, if a=I11, then a2=a, whence a is 0 or 1. But the possibility that a=0 is excluded by AV1, whence a= 1. One says that I { is trivial if jai =1 for all nonzero aeF. Every absolute value on a finite field k=Fq is trivial. This is so because for any nonzero aek, we have aq-'=I; accordingly Ia19-1=1, and hence Ia1=1, since R, has no roots of unity other than 1. DEFINITION. Two absolute values I I and I I' on F are equivalent if there is a

positive constant t such that I a I' = I a I I for all aEF. A place of F is an equivalence class of nontrivial absolute values.

Note that if we replace an absolute value I I satisfying AV-3 for some c>0 I` for some t>0, then c is replaced by c1. Appealing to the previous lemma, we see that every absolute value is equivalent to one that satisfies the triangle inequality. The next proposition is similar in both form and proof to the corresponding statement for the function modk on a local field (Proposition 4-11). by I

4.4. Places and Completions of Global Fields

157

4-28 PROPOSmoN. Let I be an absolute value on F. Then the following statements are equivalent: I

(i) 1-1 satisfies the ultrametric inequality (i.e., AV-3 with c=1).

(ii) The set {InI : nnN } is bounded. In either case, InI is in fact bounded by 1 on N.

PROOF. That the first statement implies the second follows at once from the observation that lnI =

Il+1+...+ll S 1

.

Conversely, suppose that InI is bounded by some positive constant Q for all natural numbers n. Then since I I is multiplicative, In I cannot be greater than 1 for any n, or else In-I tends to infinity. Replacing I I by an equivalent absolute value if necessary, we may assume that I I satisfies AV-3 with cS2 and hence satisfies the triangle inequality. Thus we may calculate as in the previous result:

Ia+bl"sZI(;)I

IalVIbI"-j

j=0

ImI = Imtl Im2I implies that Imd < 1 for at least one i, contradicting the minimality of m. Thus m is prime, and accordingly we shall henceforth write p for m. We claim next that Ial=1 for any integer a prime to p. Indeed, any such a is of the form dp+r for integers d and r with O 0 (S if char(F) = 0 . S u (primes P with residual characteristic * 1)

The proof will be complete once we establish the following result:

4-38 LEMMA. Suppose that P does not lie in S'. Then for any prime Q of K lying over P the local extension KQIF, is unramifred. PROOF OF LEMMA. Clearly the local extension is either trivial or cyclic of degree

1. We may thus assume the latter case, so that a is a root of f(x) in KQ-FP. Let L be the maximal unramified subextension of KQ/Fp. We have the following diagram of local and residual fields:

KgFQ

Now we consider the consequences of the relation f(a)=0. First note that whichever of the two forms that f takes, since a is a unit in D p (for P not in S, I alp=11), it follows that a is itself a unit in oQ and, in particular, integral with respect to Q. Second, f(x) has a root ill in the residual field Fq that arises from an integer of the corresponding local field. This second statement clearly holds also at the middle level of the diagram above, and this is the key to the argument. Let us compute the formal derivative of f(x):

fi(x)

11xr-'

-1

ifI=p.

170


In either case f'(f) is not congruent to zero modulo P, and Hensel's lemma (Exercise 6) applies-to the middle level!-to lift fi to an integer of L. But then L contains a root of F and KQ=L. Thus P is unramified. This complete the proof of the full proposition. REMARK. Note that the hypotheses of Hensel's lemma, namely that

(i) f(Q) = 0 (mod P) and (ii) f'(fl) 0 (mod P), imply that f and f' do not have any roots in common; that is, the discriminant off [or the resultant of (f, f')] is nonzero modulo P. This naturally leads to the use of the discriminant of K/F to determine which primes ramify-typically the more common approach.

Global and Local Bases In this subsection, K/F is a finite separable extension of global fields. Let u be a place of F,, and define M by

M=fK,, vlu

That is, M is the product of all the completions of K at places lying over u. We have an embedding w

where w,, is the canonical embedding given by the completion at v. The following result is fundamental.

be an F-basis of K, and let u be a place of

4-39 PROPOSITION. Let {eP

is an Fw-basis of M. Moreover, there exists F. Then X= ( a finite set S of places of F, containing the A rchimedean ones, such that

for all uoS, OM = fl 0Kv WIu

is free over oFu . with basis X.

4.5. Ramification and Bases

171

If L and M are extensions of a common field k, then Homk(L, M) denotes the set of embeddings of L into M that induce the identity map on k.

PROOF. Extend yi to an F-linear map

p:K®F' M in the obvious way. Both sides are F.-vector spaces of dimension n, since as we have just seen, the sum of the local degrees [K,: F.] is precisely the dimension is an F-basis for the domain. Hence it of K over F, and clearly suffices to show that q, is injective, in fact over F.. This requires one technical preliminary. Recall from our discussion of local and global degrees that K=F(a) for some a and that every embedding of K into F. over F is induced by an assignment a -a fl, where /y is a root of the minimal polynomial p(x) of a over F. Moreover, the associated place induced on K depends exactly on the conjugacy class

of /i: the assignments a N /3 and a H fi' give rise to the same place of K if and only if 6 and /y' are roots of the same irreducible component of p(x) when factored over F . The upshot of this discussion is that we can construct a bijection A. between the global and local embeddings into the algebraic closure of F. as follows:

UHomF. (K,, T.) VIM

or,

Q

(a H Q) H (Wv(a) H f )

We now proceed with the main body of the proof. Consider the following diagram:

_®1

K ®F F.

M (9 FM F.

KI

Fu HomF(K,FP)

_ ) 11°Fv(Kv,F.)

A

F u ""

1,

FHomp.(Kv,Pu)

_

vlu

where A. is the isomorphism induced from A. and K is the F. -linear injection induced by the F. -bilinear map

172

4. The Structure of Arithmetic Fields THomp(K.P,)

Kx F

(x, Y) H (a(x)Y)QEHom p(K.pu)

and similarly for each

W av factors as K K

,. Since by construction each embedding a :K -+ F. for some unique v I u, we have that

A. (K(x (& Y)) = A ((a(x)Y)o,Ha,,,

((a9 ° W,.

)V1 U

But also

and this is the same as the v-component of A..(K(x®y)). Hence the diagram is commutative, and it follows that 97 0 1 and q' are injections, as required. We now prove the second assertion of the proposition. Let u be any finite place of F. Since oF is a discrete valuation ring and therefore a principal ideal domain, each oKy is free over of , and thus oM is likewise free over op... The previous part shows that in fact the rank of oM is dimF(M)=n, the cardinality of Let S' be the finite set of places consisting of the basis X= { {v(ex), ..., the Archimedean ones, the unramifred ones, and those corresponding to primes that divide the numerators or denominators of the ej. Then X certainly lies in oM for all uvS'. Now consider the following claim:

CLAIM 1. There exists a finite set SDS' such that for all uoS, the collection { yi(e1), ...,

spans oM over oFy.

is clearly a basis for oM over oF. Granting this, the collection { y'(e),..., for all uES, as required. To establish Claim 1, we consider the modules

L=yoFe,

and

F.

.

Then for all uoS'. (The isomorphism follows from the equality of dimensions.) The claim now follows, provided that L.= oM for all but

finitely many of these u. Let P. be the unique prime ideal of oF.. Then by Nakayama's Lemma it suffices to show that

L+PuoM=oM

4.5. Ramification and Bases

173

and since u is unramified, this amounts to establishing LM+(f PV)oM =0M "I"

where P is the unique prime ideal of oK,. This in turn follows almost everywhere from our next claim.

CLAIM 2. For almost all places u, the elements yi(ej) span the product

R=fl oK/P VIM

over F=oF /PM.

Here the bar denotes canonical projection into the quotient module. To prove this, put

VIM

Then R identifies with oK/IM by the Chinese remainder theorem, since the prime

ideals P mK are all also maximal; moreover, each yi(ej) identifies with F, Thus Claim 2 is equivalent to the following, which finally we prove directly: CLAIM 3. For almost all places u, we have

PROOF OF CLAIM 3. Note that the indicated inclusion is equivalent to the state-

ment that oK c L +IM. Put 2t=LnoK. We have the following chain of equivalences:

OKQL+IM g0Ka21+IM 21+IM

some vIu some vJu

.

The second equivalence follows because IM is not contained in any maximal ideal other than P,,. But by the general theory of Dedekind domains, 21 is contained in only a finite number of prime ideals of oK, and hence we have the required inclusion for almost all u.

174


Note that since K®FFW and are both finite-dimensional of the same dimension, they acquire a canonical locally compact topological structure from F. Thus we end with the following useful result, whose proof is left as an exercise. 4-40 PROPOSITION. The algebraic isomorphism K®FF. -* M= fl,,. K. is in fact a topological isomorphism. 0

Exercises 1.

Let a be an automorphism of a locally compact group G. Show that if G is discrete, then the module of a is 1.

2.

Construct a strictly multiplicative F: N -* R, such that the conclusion of Proposition 4-10 does not hold. (In particular, F must not satisfy the given inequality.)

3.

Let V be a locally compact topological vector space over a nondiscrete locally compact field k, and let W be a subspace of V. Show the following, without appeal to the fact that V must indeed be finite-dimensional over k (cf. Proposition 4-13):

(a) V is topologically isomorphic to W®W' for some subspace W' that is topologically isomorphic to V/W. Here W and W' have the topology induced by the projection maps pra,(X) and prw,(X); i.e., the weakest topology that makes these projections continuous. (Note that both subspaces are trivially locally compact with respect to this topology.)

(b) If X is a Borel subset of V, then pra,(X) and prK (X) are Borel subsets of W and W', respectively.

(c) Let p and p' be Haar measures on W and W', respectively. Show that the product

is a Haar measure on V.

(d) Conclude that for each aek, modv(a)= modw(a) - modw,(a). 4.

Let K be a Tinitely generated extension of transcendence degree 1 of the finite field F. (Hence K is a global field.) Show that there exists an element u in K such that K is a finite separable extension of the function field F(u).

5.

Let K/F be a finite Galois extension of global fields, and let P be a prime of F. Show that G=Gal(K/F) acts transitively on the set of primes of K lying above P. [Hint: Let Q and Q' lie above P and suppose that a(Q) does not equal (and therefore is not contained in) Q' for all aEG. What, then, can

Exercises

175

one say about fIo a(Q)? Does this product not lie in P? Must it not also then lie in the prime ideal Q'?] 6.

(Hensel's Lemma) Let F be a non-Archimedean local field with ring of integers of (ac-F: I at 5 1) and prime ideal P= {xeF: Ixj < 1). Let feo,{x] be such that for some aeoF,

f (a) = 0 (mod P) but f (a) 0 0 (mod P)

.

Show that there exists beoF such that f(b)=0. Use this to show that F contains all of the qth roots of unity for q= Card(oF/P). 7.

(Krasner's Lemma) Let F be a non-Archimedean local field with algebraic closure T. Suppose that a,/3e F satisfy

jfl-ajI

where according to the characteristic of K, V(IK)=Im(I {AK) is either R,* or of the form p'oz. 5-15 THEOREM. For all global fields K, the quotient

CK = IKIK* is compact. PROOF. Recall from the proof of Theorem 5-11 that there is a compact subset (V

of AK such that AK=K+4), Since AK is locally compact, there exists a Haar measure p on AK, which we shall now fix; of course, ,u(1) is finite. Choose a compact subset Z of AK such that u(Z)>,u((D). Construct two subsets of differences and products of elements in Z as follows: ZI _ {zIZ2 : z1,Z2EZ} Z2 = {Z1Z2 : ZI,Z2EZ}

These sets are also compact by the continuity of subtraction and multiplication. Since K is discrete in AK, KnZ2 is finite, with nonzero elements, say, y,,...,yr. Now set

Ì'U6({(u,y,'v): u,v EZ1}) J=I

where 8 is the embedding of IK into AKxAK that sends x to (x,x-'). (See Exercise 1.) Since 6 is a homeomorphism onto its image, `F is a compact subset of IK, whence the theorem is a consequence of the following claim:

5.4. The Class Groups

201

CLAIM. I6 c K*`Y.

yj is the module of

PROOF OF CLAIM. First recall that for any ye IK, I Y IAK=

the automorphism of AK given by multiplication by y. Now pick anyx a Ix . Since I x IAK= 1, we see that the compact sets xZ and x 'Z have the same volume

as Z. Since p(Z) > p((D), it follows from Exercise 4 that there exist elements z1,z2,z3,z4EZ, z1xz2, z3*z4, such that a=x(z1-z2) and

8-x-'(z3 z4) are both in

K. Then aQ=(z1-z2)(z3 z4) evidently belongs to K*nZZ {y,,...,y,}. In other words, (z1-z2) (z3 z4) y;_' =1, for some j sr. Thus 5(xf)=5(Z3 Z4)=(z3 Z4, (Z1-Z2)Yj 1)EZ1XZ1yj_ 1

This shows that xfeP and completes the proof. It is useful to have S-versions of the groups we have been discussing, for any

finite set S of places of K containing Sm, the set of Archimedean places. Of course, there are no such Archimedean places if charK is positive. (This notation is unfortunately conventional, although not entirely sensible: it excludes the infinite places for a function field. Let the reader beware.) DEFINITION. Let K and SLDS,o be as above. Then define the S-ideles of K by

eo;,,Vv 0S}

IK.S

.

Equivalently,

IK,s = fK* x fl 0, ves

veS

5-16 LEMMA. 1K s is open in IK; it is compact if and only if S=0, which can occur only in positive characteristic. PROOF. That IK s is open in IK is clear, because the restricted direct product topology on IK is the same as the relative topology induced by the product (See Section 5.1.) Since K,,* is not compact for any v, 1K.S is compact if and

only if S is empty. But in characteristic zero, we require that S contain the nonempty set of Archimedean places, so this can happen only in positive characteristic, as claimed. DEFINITION. Let K and SQS. be as above. Then

I", = IK n IK.s 1

5. Adeles, Ideles, and the Class Groups

202

denotes the set of S-ideles of norm one.

According to the lemma, IK s is an open subgroup of IK in the relative topology induced by the full idele group. DEFINITION. The ring of S-integers of K is defined to be

Rs=KnAxs where AK's = (XEAK : XvEDv, VVOS) .

The definition above in particular gives oK as Rsm for K a number field, and oK as Rso for K a function field, where in the latter case So denotes the set of infinite places of K. Also note that Rs

K # n IK.s = K # n IK.s 1

.

This is because IKs is the group of invertible elements in A,,, and vas

vOS

5-17 PROPOSITION. Let S be a finite set of places of K containing.. The following statements hold: (i) The quotient IK.s / Rs is compact.

(ii) There exists an isomorphism x

Rs = pK X

Zr(S)

where pK is the group of roots of unity in K and

r(S) = sup(O, Card(S)-1 )

.

PROOF. (i) Since IK.s is open in Ii, its image IK.s/Rs is an open (hence closed) subgroup of IK / K*. But according to our previous theorem, the ambient space is compact, and hence the assertion.

(ii) Since we know this for the special case S=0 (see Exercise 5), we may assume that S is nonempty. Put


203

jxj,=1). This "adelic where the product is taken over all places and Cv (xeK,,: circle" is a compact subgroup of Ix .s. We have a short exact sequence of topological groups

-i l

1-* C -_*IK,s -*

.

YES

Note that 1.6

K* / C = R, _= R, if v is Archimedean

IZ, if v is non-Archimedean. Writing r=ri+r2, where r, is the number of Archimedean places in S and r2 is the remainder, this yields the sequence

1-*C-->IK.s ->R'' xZ'2

.

Since, again by Exercise 5, CnK* =fsK and also IK s n K* = R,' , we get the short exact sequence

1->pK-

-+L-*1

where L is the image of K* in R'' x V2. Since K* is discrete and cocompact in IK s , an application of Exercise 6 below (with A= FI shows that L is iso-

morphic to Z. REMARK. Part (i) implies in particular that RS is finitely generated as an abelian group, a fact that is not obvious from the definition. For K a number field and S=SS, this was established by Dirichlet and Minkowski. We now introduce S-versions of the idele class group, which have a critical property when S is nonempty. DEFINITION. The S-class group of K is defined by CK.S-Ix/(K*-IKS)

.

204


The critical property is this: The inclusion map of norm-1 ideles into the full idele group always induces an injection of quotients IK1K*IK,s

IKIK*IK.s

.

However, if S*O, this map is moreover an isomorphism, because we can then always represent any idele class on the right with an idele of norm one by adjusting a component corresponding to a place in S. If S is empty, then we are in characteristic p>O, and the map has cokernel isomorphic to Z by Theorem 514, part (ii). 5-18 THEOREM. The S-class groups of K have the following properties:

(i) In the case that S is nonempty, CK s is a finite group. (ii) In the case that S is empty, CK 0 is isomorphic to the direct product of Z with a finite group.

PROOF. We have seen that the image of IK,s in IK is open. Since IK /K* is compact, the quotient IK/K*IK s must then be finite. The theorem now follows U from the preceding analysis of the injection of IK/K*IK s into CK,,.

The Traditional Class Group A global field K is the field of fractions of the Dedekind domain R=DK, the ring of integers of K. A fractional ideal of K is a nonzero finitely generated Rsubmodule of K. Thus in particular, the ordinary nonzero ideals of R are fractional ideals of K. One knows from the basic theory of Dedekind domains that JK, the set of fractional ideals of K, constitutes a group under multiplication of (fractional) ideals and, moreover, that J. is a free abelian group on the prime

ideals of R. This is to say that we may write every fractional ideal aEJK uniquely as

a = 1-1 P-P P

where the product is taken over all prime ideals P of R and nP is zero for almost all P. (See Appendix B.) We sometimes write vp(a) for the exponent nf, defined

by this factorization, and similarly define vp(x) for nonzero xEK. More precisely, vp(x)=vp(xR)=ord',(x), where rr is a uniformizing parameter for R. We call vp the discrete valuation associated with P. Fractional ideals of the form Ra, aEK*, are called principal fractional ideals, and these constitute a subgroup PK of JK that includes the nonzero principal


205

ideals of R. The quotient group JKIPK is the traditional ideal class group of K, here denoted Cl,. If aeJK, then [a] denotes its projection into the class group. As previously, S. denotes the set of infinite places of a global field K. Hence SL is either S. for a number field or So for a function field.

5-19 PROPOSITION. Let K be a global field. Then there is a natural isomorphism CK.sd = C1K

PROOF. Define a map

a: 'K -+CIK

xH

[1-1 P°P(xP)} P

with vp as above. Then a is a well-defined homomorphism. Moreover, if xEK*, then

(x) = fl p'r(x) P

is the principal fractional ideal generated by x, and so a(x)=1 . Since a(x) depends only on the components of x corresponding to the finite places, a is a is trivial on nP opx, since op"cKer(vp) for all trivial on n(vEsm)Kv*. Finally,

P. In summary, a is trivial on K* 1Ksm and hence induces a homomorphism a : CK.sm -+ CIK

sending the class of x to a(x) . Suppose that aeJK. Then vp(a) is nonzero for only a finite number of P. Accordingly, we may define an idele x by requiring that x be nonzero at the infinite places and xp= .1;p+P(a) for the places corresponding to primes P, where ,r, is

the associated unifonmizing parameter. Then by construction a([x])=[a], and thus a is surjective. Finally, suppose that a([x])=1 for some xelK. Then there is a yEK* such that (Y) _ r-1 P"P(xP) P

This implies that for all P, Vp(y)=Vp(xp), and so we may choose u=(up) a flop' such that (xu)P yp, for all P. Then xu and y differ by an element of fl(VEs )Kv*; that is, x and y differ by an element of 'K3 Consequently, xEK* IK s., which means that its class [x] in CK s is trivial. Hence a is also injective. 0


206

REMARK. By saying that a is natural we mean that it is functorial for the inclusion of fields in one direction and for the norm map in the other.

Ray Class Groups with fractional Again let K be a global field, the fraction field of ideal group JK. Let M be a nonzero integral ideal of R, so that we may factor M uniquely as

where P, is the prime corresponding to the finite place v of K, with associated discrete valuation vp. Let S be the set of finite places where vp(M)>0. DEFINITION. An element aEK* is said to be congruent to 1 mod M if the following conditions hold at every V ES:

(i)

aE o;,

(ii) Vp(a-1)ZV (M) The set of all such a is denoted KMi; one checks easily that this constitutes a subgroup of K*. DEFINITION. Let K and Mbe as above. Then define

JK(M) _ {IEJK : (I,M) = R}

.

That is, JK(M) consists of the fractional ideals of K that are comaximal with respect to M. In particular, if then aREJ(M). We may thus further define ClK(M) = JK(M) /KM

I

We call ClK(M) the (wide) ray class group of K relative to M (or with conductor M).

EXAMPLE. Consider the case K=Q. Then R=Z is a principal ideal domain, R" = (±1), and each nonzero integral ideal M takes the form mZ for some unique positive integer m. Define a map

V : C1K(M) --4 (Z/mZ)'/{±l}


207

that sends the class of a fractional ideal (a/b)Z (with both numerator and denominator prime to m) to the double residue class ±[a][b]-l. This map is welldefined on JJ(M) and factors through Clx(M) because (a/b)Z maps to the identity if and only if a m±b (mod m) in the elementary sense. Since 9 is clearly surjective, it is in fact and isomorphism. More generally, for a number field K this construction is usually extended to include the signs at the real places. Let {w1,...,w1) be a set of real embeddings representing inequivalent real places (not necessarily exhaustive), and put

M=(M,w,,...,w,) where M is an integral ideal.

is said to be congruent to 1 mod M if the fol-

DEFINITION. An element

lowing conditions hold at every veS: (i)

an I (mod M), as above

(ii) w(a)> 0, t/j=1,....I The set of all such a is denoted KM 1 , and as previously, this constitutes a subgroup of K*. DEFINITION. We define the quotient

Clx(M) = J., (M)/KM, When {w,, ...,w,} comprises the entire set of real places of K, then this is called the narrow ray class group of K relative to M. EXAMPLE CONTINUED. Again consider the case K=Q. Let M = (M,00), with M generated by m>0 as before. We can now in a sense refine our map rp to an isomorphism

Clx(M) = (Z/mZ)" The point is that by using the narrow ray class group, we can distinguish signs in (Z/mZ)". More particularly, given any ideal xZ in JK(M), we take x=alb, with a and b uniquely given positive integers relatively prime to m and to each other, and then map xZ to [aJ[b]-l. This map is clearly a surjective homomorphism with kernel KM ,

K.

208


Exercises 1.

Let K be a global field, and let A' have the product topology. Show that the mapping

Ix -* A6 x i-- (x,x-I) is a topological isomorphism onto its image (under the relative topology induced by that of the codomain). 2.

Let A be an integral domain for which all prime ideals are maximal. Show that if PI and P2 are distinct prime ideals of A, then +P2 = A

for all positive integers m and n. [Hint: Prove this directly for all m when n=1, and then proceed by induction.) 3.

Let K be a global field. Use the discrete embedding of K into the associated adele group and Exercise 1 to show that K* embeds discretely in the associated adele group.

4.

Let G be a locally compact abelian group with Haar measure u. Suppose that F is a subgroup of G and that 4) is a compact subset of G such that G=F+(D. Show that if X is a compact subset of G such that u(X)>p((D), then there exist distinct elements xi,x2EXsuch that

5.

Let K be a global field. Show that IxL= 1 at every place v of K if and only if x is a root of unity in K.

6.

Let G be a topological group isomorphic to R'x Z'+'-' for some integers sz rZO, and let A:G->R be a nontrivial, continuous homomorphism such that when r>0, A is in particular nontrivial on W. Assume that IF is a discrete, cocompact subgroup of Ker(A). Show that 1--=Z'.

7.

Let K be a global field. Show that the isomorphism a: CK s ral in the sense of the remark following Proposition 5-19.

8.

Let K be a global field and let S be a finite, nonempty set of places of K containing the infinite ones. Show that RS [=KnAKS), the ring of Sintegers of K, is a Dedekind domain. [Hint: Appeal to the case S=S,,, where we know this to be true by Appendix B.]

= C!K is natu-

Exercises

9.

209

For any Dedekind domain R with fraction field K, define Pic(R) to be the group of invertible fractional ideals of K modulo the principal ones. With this definition and the preceding exercise in mind, prove the following Sversion of Proposition 5-19:

Let K be a global field, and let S be a finite nonempty set of places of K containing the infinite ones. Then there is an isomorphism CK s = Pic(Rs). Show also that for S large enough, CK s is trivial.

10. Let K be a number field. (a) Show that an element xeK* is a unit of oK if and only if NK,Q(x)=±1.

Assume for the remainder of this exercise that K is a quadratic number field; that is, K=Q(S), where Sz=d, a square-free integer. (b) Show that Z[5) OK

if d = 2,3 (mod 4)

Z[l2S[ if das 1(mod 4).

(c) Assuming that d is negative, list the units of oK.

(d) Assuming that d is positive and congruent to either 2 or 3 modulo 4, show that the units of oK are precisely those numbers a+bS such that the integer pair (a,b) satisfies Pell's equation a2 - db2 = ±1. Show, moreover, that there is a fundamental unit u,=ai+b,5, a,,b,>0, such that every unit in oK is of the form ±u1" for some nEZ. The pair (ai,b,) is called a fundamental solution to Pell's equation.

(e) For this part, we assume that the reader is familiar with continued frac-

tions. Assume that d is as in the previous part, and let [ao,be the (simple) continued fraction expansion of S with "convergents"

A"IB" _ [ao,a1,...,a"]

.

Show that for some n, the pair (A",B") constitutes a fundamental solution to Pell's equation. Check in particular that when d=2 (respectively, 3), the fundamental unit of K= Q(S) is 1+.5 (respectively, 5 +25).

210


11. Let R be a Dedekind domain, for example the ring of integers in a number field.

(a) Suppose that Pic(R), the class group of the fraction field of R, is trivial. Show then that R is a unique factorization domain; that is, every nonzero element of R is expressible as the (finite) product of irreducible elements and that this factorization is unique up to order and associates. [Hint: Show that any two elements a,bER, not both zero, have a greatest common divisor by looking at the intersection of aR with bR.] (b)

Show that every integral ideal I in R can be written as the intersection of a finite number of principal ideals.

(c) Prove the converse of part (a): if R is a unique factorization domain, then Pic(R) is trivial. [Hint: To show that every integral ideal is principal, show first that having unique factorization forces the intersection of any two principal ideals to be principal, and then appeal to part (b).] MORAL. The class number hK of a number field measures the failure of unique factorization in oK.

12. (Artin) This exercise develops an explicit description of the connected component of C. Let K be a number field of degree n=r,+2r2, where r, and r2 are, respectively, the number of real and nonconjugate complex em-

beddings of K into C. Recalling that oK has rank r=r,+r2-1, fix a set {uP..., ur) of multiplicatively independent units in oK. Put

V=R®Z and embed Z in V by the diagonal map that sends m to (m, m). Write IK = IK X If

where the elements in IK (respectively, IK) have only trivial finite (respectively, infinite) components.

(a) Show that for any y e If and x e Z, the expression yx makes sense. [Hint:

K of neighborhoods of unity consisting of subIf has a fundamental system K

groups of finite index.]

(b) Show that for any z E If and t E R, the expression z' makes sense, and that K it can be normalized to obtain real values at real places.

Exercises

211

(c) For j= 1, ..., r2 and to R, let 4 (t) denote the idele with component a 29;t at the jth complex place and 1 everywhere else. Define a mapA by

V' ®R'2 -+

A

I, r

r

flu,'2'OJ(tj)

(A,t)= (A,,...,Ar,t1,...tr2) H 1=1

J=1

Show that A(A, t) is a principal idele if and only if every A, and every t. lies in Z.

(d) Let A : V' ® R'2 -+ C'' denote the induced map to the idele class group. Show that V/Z is compact, connected, and infinitely and uniquely divisible. Conclude that D=Im(A) is compact, connected, and infinitely divisible.

(e) Show that every infinitely divisible element of C'' lies in the closure of D, and hence lies in D itself.

(f) Show that D contains the connected component of CK, and conclude that in fact D is the connected component of C'' . [Hint: Use that D contains the image of IK mIf .J

13. Let R be a commutative ring with unity. Define the Heisenberg group of R as follows:

a

1

H(R) =

0 0

b

1

c :a,b,ccR

0

1

Show that for any global field, H(K) embeds as a discrete, cocompact subgroup of H(AK).

14. Continuing in the context of the previous problem, show that the abelianization map

H(AK) -1AK 1

a

b

0

1

c

0

0

1

induces a continuous surjective map

H (a, b)

212


fr : H(AK)/H(K) -> (AK/K)2

whose fibers identify with AK/K.

15. Let K be a global field. (a) Show that GL2(K) embeds as a discrete subgroup of GL2(AK).

(b) Show that the corresponding quotient space-not quotient group, for this embedding is not normal-is not compact.

(c) Let

denote the subgroup of consisting of scalar matrices. Show that the quotient space GL2(AK)/Z(AK)GL2(K) is still not compact.

6 A Quick Tour of Class Field Theory

One could argue that the principal goal of number theory is to understand the integral or rational solutions of systems of Diophantine equations; that is, polynomial equations with integral coefficients. Nineteenth-century mathematicians, mainly riding the impetus provided by attempts to tackle the Fermat equation x"+y"=z" (n23), realized the benefits of studying the solutions in extended number systems R, as opposed to confining one's attention to only Z and Q, and this led eventually to global and local fields and their rings of integers. Such an extension often was made to allow for the presence of suitable roots of unity in R, which provided desirable factorizations, such as

y)

X. +y" _ j=o

Two related problems immediately arose, the first associated with the general failure of unique factorization in R, leading to the class group, and the second pertaining to the question of how rational primes factor, or split, in R. The latter problem was first solved in its entirety, in the guise of the study of quadratic

forms, for quadratic fields F=Q(5), where 6-2=D is an integer that is not a square in Q. It was established that an odd prime p splits in F if and only if D is a quadratic residue-that is, a square-mod p, and that the set XD of primes for which D is a quadratic residue mod p completely determines the extension F/Q. (In modern parlance, one says that the set XD defines a canonical open subgroup of the idele class group CQ.) Of special importance here is the quadratic reciprocity law, which for primes p and q gives a precise relationship between the status of p as a quadratic residue mod q and the status of q as a quadratic residue modp. Further progress followed on cyclotomic and Kummer extensions, and, perhaps most significantly, an assertion of Kronecker led to the realization of all abelian extensions of Q as subextensions of the cyclotomic ones.

By the early twentieth century, the central problem of algebraic number theory had become that of describing the splitting of primes in finite abelian extensions ("class fields") L of an arbitrary number field K in terms of structures associated with K itself. A particular subclass that was well understood

214

6. A Quick Tour of Class Field Theory

early on was the maximal unramified abelian extension 11(K), called the Hilbert class field of K, whose Galois group Gal(H(K)/K) turned out to be isomorphic to the ideal class group Clx. In the 1930s, Takagi gave a general solution to the problem and established in the process an abstract isomorphism of the Galois group of any finite abelian extension L of K with a ray class group of K. (As we have seen in Chapter 5, every ray class group is a quotient of the idele class group.) A completely satisfactory understanding of abelian extensions LIK was finally achieved with the revolutionary work of E. Artin, who proved a general reciprocity law. Artin reciprocity, on the one hand, vastly extends Gauss's law of quadratic reciprocity and, on the other, gives a canonical isomorphism between Gal(L/K) and the relevant ray class group. The key tool is a crucial homomorphism called the Artin map. In this chapter, after introducing the required technical preliminaries on Frobenius elements, the Tchebotarev density theorem-a huge generalization of

Dirichlet's theorem on primes in arithmetic progressions-and the transfer map, we summarize (without proof) the main results of abelian class field theory i1 la Artin. While we state everything for idele class groups rather than ray class groups, the reader may consult Section 5.4 for the relevant dictionary. Putting matters in adelic language might seem an unnecessary complication,

but it is absolutely essential if we are to apply the techniques of harmonic analysis. We end the chapter with an explicit description of the abelian extensions of Q and Q. including a proof of the Kronecker-Weber theorem. SPECIAL NoTEs. (i) The results of this chapter are not prerequisite for the proof of Tate's thesis in the following chapter, but they will play a role in some of our applications. (ii) In the exercises for Chapter 7, we shall develop a proof of the

Tchebotarev density theorem as reformulated in terms of Dirichlet density. Since this proof in fact relies on Artin reciprocity, it is important to stress here that Artin's law is itself independent of the Tchebotarev density theorem. While Artin was inspired by ideas in Tchebotarev's proof, his actual argument does not depend upon it, and hence we introduce no latent circularities.

6.1 Frobenius Elements The goal of this section is to introduce a family of special elements-or, more properly, of special conjugacy classes-in the Galois groups of global fields. We fix a global field F, and for any Galois extension K/F denote the corresponding Galois group Gal()M. .

We shall first consider the case of a finite Galois extension K/F with G=Gal(K/F). Let Q be a prime of ox. Then Q lies above some prime P in o.,, and we let F denote the residue field o,/P. Recall from Section 4.3 that we then define the decomposition group of Q in G to be

6.1. Frobenius Elements

215

DQ= tar= G: a(Q)=Q}. Let the residue field oK/Q be identified with the finite field Fq. Then we have a canonical homomorphism PQ : DQ -+ Gal(Fq/F)

that associates with aeDQ the map (x modQ i-4 a(x) modQ) for all xEOK. As proven previously, the map pQ is always surjective and is in fact an isomorphism if and only if P is unramified in K. Moreover, each aE DQ extends to an automorphism of the completion KQ that is trivial on the subfield FP; the induced map

jQ:DQ-Gal(KQ/FP) is unconditionally an isomorphism. One knows from elementary field theory that Gal(Fq/F) is cyclic, generated by the Frobenius map

XHX'f where Card(F)=pf. With this in mind, we make the following definition.

DEFINmoN. Let P be unramifred in K. Then the Frobenius element gPQ,P in DQCG associated with QIP is defined by = PQ (x H x'f)

Note that this element unfortunately depends on the choice of Q over P. Indeed, suppose that Q' is another prime dividing P. Note first that DQ, is conjugate to DQ. Explicitly, we know that we can find fEG such that f(Q')=Q, and consequently 8-' aji preserves Q' for each aEDQ. 6-1 LEMMA. The maps ipQ,p and cQTP are conjugate in the Galois group G.

PROOF. Choose f EG as above so that flDQ,,O-' = DQ. Then by definition,

PQ,,, =PQ (XHX'f)

.

To show that this is conjugate by fl to q.'Q1p, we explicitly compute as follows:

216

6. A Quick Tour of Class Field Theory ,

'-pQ/PN(x+Q')=/' "cqQ/P(Nx)+Q)

=f'(i(xp,)+Q) =xpf+Q' =PQ'/p(x+Q') This completes the proof.

DEFINITION. The Frobenius class in G corresponding to P, denoted q'P K/F or (P, K/F), is the conjugacy class of the Frobenius element op,,P.

This is well-defined by the previous lemma. The notation (P, K/F) is sometimes called the Artin symbol of P relative to K/F We shall next analyze the functorial properties of these Frobenius classes. 6-2 PROPOSITION. The Artin symbol has the following properties: (i) Let M/F be a finite Galois extension and K/F a normal subextension,

so that the restriction map NM/K from Gal(M/F) to Gal(K/F) induces an isomorphism between Gal(M/F)/Gal(M/K) and Gal(K/F). Then for any prime P unramifred in M, NM/K(P,MIF) = (P,K/F)

.

(ii) Let K and K' be two finite Galois extensions of F that are, moreover, linearly disjoint over F. Then for every prime P unramifred in KK; we have that anGal(KK'/F) lies in the Frobenius class (P,KK'/F) if and only if(alK,olK,)E(P,KIP) x(P,K'/F).

(iii) Let K/F be a finite Galois extension, and let L be an intermediate field, not necessarily normal over F, with [K:L[=m. Let P be a prime of F unramified in K, and suppose that Q is a prime of L that divides P and that P is a prime of K that divides Q. Then we have that LQ=Fp if and only if OpP,P E Gal(K/L). Moreover, the number of primes Q of L lying over P such that LQ FP is given by the formula

1m Card ({a E Gal(K/F) : aipi,,p a-' E Gal(K/L) }) REMARK. A prime Q of L dividing P such that LQ=Fp is called a degree-one prime over F. When LIF is normal, (P,K/F) is a subset of Gal(K/L) if and only

6.1. Frobenius Elements

217

if p,,p eGal(K/L) for some Frobenius element defined over P, and in this case P splits completely in L into a product of primes of degree 1. PROOF. (i) Let P' be a prime of K above P, and P" a prime of M above P'. Let k, k', and k" denote the corresponding residue fields. Then we have the following diagram:

M --t

Mp. LD op.,

-* k"

I

K - Kp,

I

o p,

-1

k'

op

--

k

I

I

F ->

Fp

Let k=F4o, and let Q'eGal(k'/k) and or"eGal(k"/k) denote, respectively, the Frobenius automorphisms of k' and k" over k. Both Q' and or" are given by the assignment x H x'°, and thus it is clear that or' is no more than the restriction

of a". Moreover, since P is unramified in M, the decomposition groups Gal(M/F) and Dp,c Gal(K/F) are, respectively, isomorphic to Gal(k"/k) and Gal(k'/k). Since by construction pp,,,peGal(M/F) and pp,,peGal(K/F) are the preimages of ar' and a" under these isomorphisms, we see also that ip,,,p is the restriction of opp,,,p. The same then holds for the associated conjugacy classes, and hence (i) holds. (ii) Let

r: Gal(KK'/F)-*Gal(K/F)xGal(K'/F) dH(01K10IK') denote the canonical homomorphism. This is in fact an isomorphism because K and K' are assumed linearly disjoint over F. Now let P denote a prime of KK' lying above P. Then Q = T n oK and Q' = P n OK, are, respectively, primes of K and K' lying under P and over P. One checks easily that P) IPI K = c'QIP and 9PIpI K' _'Q'Ip

Conversely, any pair of intermediate-level Frobenius maps must arise via y from a conjugate of VPp because y is an isomorphism. This proves (ii). (iii) Let the primes P, Q, and P be as shown:

218


K

Ox

I

I

L

oL

I

I

F

I

QQ I

of Q P

Again we must keep in mind two elementary, but crucial facts: the Frobenius map r = Vp,p lies in the decomposition group Dp cGal(K/F), and Dp =_ Gal(K),/Fp)

where this isomorphism is nothing more than extension of an automorphism of K over F to one of K¢ over Fp. Now if Q is in fact a degree-one prime, which is to say that LQ=Fp, then the corresponding extension of r is ipso facto trivial on

LQ, and therefore on L. Thus reGal(K/L). Conversely, if rEGal(K/L) and qo=Card(ofJP), then it follows that aqua (mod Q) for aeoL, from which we deduce at once that the residue fields of L and F are identical. Accordingly L,Fp, as required. This proves the first statement of (iii). To conclude, we establish the formula. We know now that the number of primes P of K dividing P such that P n oL is of degree one over P is exactly the number of Frobenius elements defined over P that lie in Gal(K/L); this is just Card((P,K/F)nGal(K/L)). Now the number of such primes lying over any single given degree-one prime in L is always m/f, where f be the residual degree associated with T. (Clearly f is the same whether computed with respect to L or F and is therefore independent of P n oL, provided that this intermediate prime is indeed of degree one.) Thus the number of degree-one primes in L is f/m times the cardinality of (P,K/F) n Gal(K/L). But every element of this intersection is represented exactly f times in the form o qF,,, 0'-' as Q runs over Gal(K/F), and from this the formula follows at once. O

Arbitrary Unramifred Extensions Recall that an extension OF is called unramified at a place u of F if there exists a chain

of finite extensions such that each E,/E,._, is unramified (in particular, finite and separable) at every place of E, 1 lying above u.

6.2. The Tchebotarev Density Theorem

219

DEFINmoN. Let F be a global field and P a prime in F. Then PW(P) denotes the maximal subextension of F/F that is unramified at P. This is called the maximal unramified extension of F at P.

It is easy to check that F°(P) exists (see Exercise 1 below) and is a Galois extension of F. For we have seen that each step in the tower that defines an unramified extension is the splitting field of a polynomial of the form x'-I, and hence is itself Galois over F. Clearly,

Gal(F' (P)/ F) = lim K / F 4-

where K runs over finite Galois extensions of F contained in F(P). The previous proposition shows that the Frobenius classes cK,F (P,K/F) patch nicely to give a class (P, F(P) IF) in Gal(F(P)IF). It is perhaps disappointing that we cannot define the Frobenius class in the absolute Galois group Gal(F/F) , but we point out without proof that if

p :Gal(F/F) -+ GL (Q,) is a continuous representation arising from the /-adic cohomology of a smooth projective variety over F (with I a prime different from the characteristic of F), then p is unramified at all P outside a finite set S of primes. In other words, p factors through GF S, the Galois group of the maximal extension of F in k that is unramified outside S. Since Gal(F(P)/F) maps onto GF5, we see that p(rep) is well-defined at every PoS.

6.2 The Tchebotarev Density Theorem Given a finite Galois extension K/F of global fields, we have seen how to define a map Gal(K/F)N V=' KIF: EF - SKIF PHop P

where E. denotes the set of places of F, SK/ , denotes the (finite) union of the Archimedean places and the finite places that ramify in K, and Gal(K/F)# is the space of conjugacy classes of Gal(K/F). A natural question to ask is whether every conjugacy class is opp for some P. The answer is yes, as affirmed by the following beautiful result, given here without proof.

220


6-3 THEOREM. (Tchebotarev) Let G=Gal(K/F). Then for every conjugacy class

C in G there exist infinitely many primes P such that opp=C. More precisely,

lira =gym

Card{P: N(P) 5 x, q = C} Card{P:N(P) s x}

- Card(C) Card(G)

Here, N(P), the (absolute) norm of the prime P, is the cardinality of the associated residue field. The limit on the left side of the equality is called the natural density of the set described in the numerator. (As noted above, we prove a reformulation of this theorem in terms of Dirichlet density in the exercises for Chapter 7.)

An illuminating special case of this theorem arises when F=Q and K=Q(Q, the field of mth roots of unity over Q, for some m> 1. Then one knows that the Galois group G of K/F is abelian, and in fact isomorphic to (Z/mZ)". Explicitly, each a relatively prime to in gives rise to an element // a

°a' Cm H `gy

of G. For every prime p not dividing in, this extension is unramified (proved for

in prime in Section 4.3). Now let C be a conjugacy class in G, so that in the present case C corresponds to a singleton subset {a}c(Z/mZ)". Then one can deduce that

q7v={Qa} a paa(mod m). Thus Tchebotarev's theorem becomes the well-known theorem of Dirichlet on primes in arithmetic progressions, namely that there are infinitely many primes p congruent to a modulo in, and, more specifically, that the density of such primes is 1/q(m).

6.3 The Transfer Map In preparation for the statement of the Artin reciprocity law, we now introduce a subtle and entirely group-theoretic construction that is of interest in its own right. The subtlety lies in that in general there is no homomorphism from a group to a subgroup. Let G be any group, with H a subgroup of finite index. Let (G, G) denote the commutator subgroup; i.e., the subgroup generated by the products sts 1 t_1 where s and t vary over G. Since conjugation by any element is an automorphism of G, the commutator subgroup is normal in G, and the corresponding quotient group Gb=G/(G,G) is called the abelianization of G. The homomor-

6.3. The Transfer Map

221

phism in question is the transfer map, also called die Verlagerung by German aficionados, V: G'b->H'b

and is defined as follows.

First choose a section s:H\G-+G; that is, a set of representatives for H\G, the set of right cosets of H in G. Put hx.Y = s(Y)xs(Yx)-' e H

where Yx denotes the effect of right translation on a coset Y in H\G. (Of course,

H\G is a G-setl) Clearly, h,.y measures the failure of s(Y)x to equal s(px); that is, the failure of s to be a G-map. Next define

V(x)= fj hry mod(H,H) Thus the right-hand side is the natural image of the given product in H'b. 6-4 PROPOSrrtON. The map V : G-+Hb is a group homomorphism independent of the choice of the section s:G-,H\G.

PROOF. First we show that V is independent of the choice of section. Let s' be another section. Then there is a function rl:H\G-+H such that s'(Y) = rXY)s(Y)

for each

eH\G. Given XE G, the direct calculation

fls'(Y)xc'{px)-' _ f Y

Y

F1 rXY)s(p)xs(Yx)-'

rl(y)-1

shows that we may calculate 17(x) using either section and obtain the same results modulo the commutator subgroup (H, H). We may make a similar calculation to see that V is a group homomorphism:

222


V(x,x2)=fls(y)x,x2s(yx,x2)-' mod(H,H) Y

= fl s(Y)Xis(YX1)-"S(Yx,)x2s(YXi x2)-' mod(H, H) Y

= f s(y)xxs(Yx, )-' . fl s(yxi )x2s(YX1x2)-' mod(H, H) P

Y

=V(x,)V(x2) In moving from the second line to the third, we note that the indicated trios of

factors all lie in H, whence all of the right hand-trios can be accumulated modulo (H,H) into a single product. In moving from the third to the fourth, we note that as y varies over H\ G, so does yx, . In consequence of this proposition, it follows from the universal property of the abelianization of a group that V induces a unique map V:

which we call the transfer map. We also write Vo H for this map to emphasize the domain and codomain. From the previous proposition it follows that the transfer map is completely intrinsic to G and H, and independent of the choice of section. Moreover, it satisfies a kind of transitivity: 6-5 PROPOSITION. (Transitivity of the Transfer Map) If HcKcG, then VG.H = VK.H c Va.K

PROOF. Exercise.

In. his book The Theory of Groups, M. Hall gives an alternative development of the transfer map via monomial representations (1959, pp. 201-203).

6.4 Artin's Reciprocity Law One of the major success stories in number theory this century has been the work of Takagi and Artin on the description of abelian extensions of global fields. This is codified elegantly and concisely by the Artin reciprocity law. In this section we shall, without proof, state this law simultaneously for global and local fields and indicate its associated functorial properties. We begin with a few preliminary considerations. Let Fbe a global or local field. Put

C., -

if F is local (F* IF IF * if r is groom.

6.4. Artin's Reciprocity Law

223

We know that this is a locally compact abclian group. Moreover, if K/F is a finite extension, we will be concerned with two natural homomorphisms:

,jwF : CF -- CK and NK,F : CK -i CF .

The first map is simply that induced by inclusion. The second is the norm homomorphism, which in the global case is induced by

sir

F. is the ordinary norm. Observe that this idele class version of the norm is well-defined: the ordinary version maps integers to integers (cf. Appendix B, Section 2), and Proposition 4-39 shows that elements of K map to elements of F. Note also that according to Exercise 3 below, the image of NKiF where

is an open subgroup of CF. Next fix a separable algebraic closure F and put

rK= Gal(F/K) for any extension K/F with K c F. To describe the functoriality of Artin reciprocity, we shall also need two maps on the Galois side. The first is simply the inclusion

iK/F : rK -- rF .

The second, which goes in the opposite direction, is the more subtle transfer map V:

r"->rKb

defined as above on the abclianizations of the domain and codomain.

Before stating Artin's reciprocity law, let us take note of the relationship between the cokernel of the norm map and the Galois group for four particular extensions K/F. CASE 1. Let F= R and K= C. Then Gal(C/R) _ { 1, p}, where p denotes complex conjugation. Moreover, the cokernel of the norm map Nc,R : C* --), R*

is simply the quotient of R* by the nonzero squares, which is to say the cyclic

group of order 2. Hence there is a unique abstract isomorphism between R*/N(C*) and Gal(C/R) sending the class of -1 top.

224


CASE 2. Let F= C. Then since the complex numbers are algebraically closed, K must also equal C, and both the cokernel of the norm map and the Galois group are trivial.

CASE 3. Let p be an odd prime, and let F= Q. and K=Q,(i), where 52=2. Then NKiF(K*) = {re Q,*: r = x2-2y2, for some x, yeQQ} .

It is a good exercise to check that this norm subgroup has index 2 in Q;. Of course, Gal(K/F) is also cyclic of order 2.

CASE 4. Let F=Fq, and let K be any finite extension. Note that the norm map from K* to F4 is always surjective, and hence has trivial cokernel. Hence the situation here is very different from that of a local or global field. 6-6 THEOREM. (Artin Reciprocity) Let F he a global field or a local field. Then there exists a homomorphism, called the Artin map,

BF : CF -+ Gal(F/F)'b = r'F satisfying each of the following two groups of assertions: PART ONE-The Artin Map for Abelian Extensions (i) For every finite abelian extension K/F, let BK/F denote the composition

of OF with the natural projection Fb -> Gal(K/F). Then BK,F is surjective with kernel NKIF(CK).

(ii) Conversely, given any open subgroup N of CF of finite index, there exists a finite abelian extension K/F such that N=Ker(9K,F). In particular,

CF /N = Gal(K/F) .

(iii) Let K/F be a finite abelian and unramified extension of the nonArchimedean local field F with residual extension k'/k. Then we have explicitly OK/F(x) =

W{z)

where to is the Frobenius element of Gal(K/F)aGal(k'/k)=(p).

6.4. Actin's Reciprocity Law

225

(iv) Let K/F be a finite abelian extension of global fields, and let P be a finite prime in F that is unramifted in K. Denote by xP the class in CF defined by the idele

T place P

all of whose components are 1 except at the place defined by P, where the component is a uniformizing parameter jr. Then we have OK/F(xp) = VP

where qP=(P,K/F). [Note that since K/F is abelian, the Frobenius conjugacy class cP is in fact a single element of Gal(K/F).]

PART Two-Functoriality

Let K/F be a finite separable extension, not necessarily abelian (with F either global or local). Then we have the following two commutative diagrams: (i)

oK

CK

NK/F j.

,J.

CF

'K /F

IF

(ii) oK

CK 1K/F T CF

Moreover, if K1/F Is an abelian extension with sub extension KIF, then we have a further commutative diagram

226


F*/Nx,/F(K*)

Ox,/F'

Gal(K1/F) proj

proj

Gal(K/F)

F*/Nx,F(K*) Bx/F

Note well that the inclusion-induced map on the class group side corresponds to the transfer map on the Galois side and that the inclusion map on the Galois side corresponds to the norm map on the class group side. In the abelian case, we may simply identify the projections.

6.5 Abelian Extensions of Q and Qp In this section, working over either Q or Q. we consider class field theory in a particular and concrete setting. We prepare with some general field-theoretic notions.

Let F be a field, for which we implicitly fix an algebraic closure. If K, and K2 are Galois extensions of F, then so is their compositum K1K2, and in fact we have an embedding Gal(K1K2/F) - Gal(K1/F) x Gal(K2/F)

Q H ((71x,,aIK2) which is an isomorphism if K, and K2 have intersection F. Thus if K, and K2 are moreover abelian extensions, so again is their compositum. Thus there exists a maximal abelian extension F"b of F, which is precisely the compositum of all abelian extensions of F within its algebraic closure. Henceforth, for any n?1, F,, denotes the field obtained by adjoining the nth roots of unity to F (again, within its fixed algebraic closure). We shall soon see that this is always a finite Galois extension of F. We further let F. denote the compositum of all of the F,,, nzl. We now state the main theorems of this section.

6-7 THEOREM. Let F be a local or global field. Then for all n, F. is a finite abelian extension. Moreover, the following assertions hold: (i)

If F=Q, then

by an isomorphism that associates ae(Z/nZ)" with the automorphism of F,, induced by w N w°, where w is a primitive nth root of unity. Consequently, F.cF°b.

6.5. Abelian Extensions of Q and Qp 227

(ii) IfF=Q. and n is relatively prime to p, then F,, IF is unramifred, with Gal(F,,/F) cyclic. In fact, every finite unramifred extension of Qp occurs as some F,,, with (p,n)=1.

(iii) If F= Q. and n is a power of p, then F,,IF is totally ramified, with Gal(F"/F) =(Z/nZ)". 6-8 THEOREM. (Kronecker-Weber) Let F be either Q or Qp. Then F.=Fb.

REMARK. Let F=Q. Then by the Kronecker-Weber theorem, given any finite abelian extension K of F, we can find a positive integer n such that K is contained in the field F,,=Q(ezirV"). Thus one can think of K as being generated by the values of the function ell's at rational arguments. Kronecker's Jugendtraum (youthful dream) was to hope that any finite abelian extension of a number field F could be generated by values at algebraic arguments of a suitably chosen set of transcendental functions. This dream is realized for imaginary quadratic fields F, where the abelian extensions are all generated by the values of elliptic functions at "division points." Further progress has been made by Shimura and others. Kronecker's dream has in fact influenced much of modern number theory.

PROOF OF THEOREM 6-7. We begin with some basic Galois theory. Let F be any

field with separable algebraic closure F. For positive n, consider the equation

Ax) = x"-1 over F. Then its splitting field is precisely F". If char(F)=p>O, then there are no nontrivial p-power roots of unity in F. Thus if we write n=p'm, with in prime to p, the nth roots of unity in F are the same as the mth roots of unity. Therefore, in the case of positive characteristic p, we can and shall assume that n is prime top. _ Let w be a primitive nth root of unity in F, so that w"=1, but w'"# 1 for any positive in smaller than n. Indeed, such an w must exist because the formal derivative off f'(x) = nx"-1 is nonzero for nonzero x-after all, n is assumed prime top in positive characteristic-and therefore f must have n distinct roots in F. Hence this necessarily cyclic group of solutions must have order n and, of course, a generator co. From this we see at once that F,,=F(w) and that F,, is the splitting field of a separable polynomial over F. Accordingly, F,, is a finite Galois extension of F. Fix a primitive nth root of unity as Then of is again a primitive nth root of unity if and only if r is an integer prime to n. In this way we obtain exactly

228


it must send W to another prim-

q(n) primitive roots. Now if itive nth root of unity. Thus we must have a(W) = Waa

for some aae (Z/nZ)". Moreover, for Or, rE G, Waar = (Qr)(W) = ,(War) = Waaar

.

Thus aar agar, and so we have a homomorphism of groups y: G -+ (Z/nZ)'

Since ais the identity of G if and only if Waa is w itself, which is to say, if and only if as=1 in (Z/nZ)", it follows that y is injective and that G is abelian. Keeping in mind that w generates F,, over F, each element ae(Z/nZ)" conversely gives rise to an automorphism or. of F,, defined by Qa(W) = Wa .

However, this might.not be an element of G by virtue of its failure to restrict to the identity on F, and this will indeed occur if some power of w lying in F is moved by a.. Hence in general y is not surjective. Before specializing to Q or Q. we observe that for any din, we may define a factorfd(x) off(x) by setting H(X-CO bn/d

.fd(X)=

br,(Z/dZ)"

Then clearly,

f(X) = JJ fd (X) dIw

with fi(x)= (x- 1). We customarily call fn the nth cyclotomic polynomial. Since

J,%.X)

fd(X)

djn.dxn

we see inductively that each cyclotomic polynomial lies in F[x[. Moreover, since each is monic, it follows from the Euclidean algorithm that its coefficients in fact lie in the subring of F generated by 1. In this sense, the cyclotomic poly-

6.5. Abelian Extensions of Q and Q,

229

nomials are generic over any field, although they may or may not be irreducible depending upon F.

With these preliminaries in hand, let us now proceed to each of the three statements of the first theorem.

(i) Now let F= Q. Then for nz3, F contains no nth root of unity, and therefore w° does not lie in F for any a prime to n. Moreover, F. is simply the splitting field of the cyclotomic polynomial f, and elements of G permute the primitive roots of unity. Thus to show that y is an isomorphism, it suffices to show that f,,(x) is irreducible, for then the order of G will be the degree of f,,, which is clearly q(n), the order of (Z/nZ)'. Let g(x) be the irreducible factor of f (x) that is the minimal polynomial of w over F. Since g is the product of linear factors of the form (x-w°), its coefficients are both rational and integral over Q, which is to say that g(x)eZ[x]. We claim that it is enough to show that for every prime p not dividing n, a)p is also a root of g. For this implies by iteration that ae is a root of g for all a prime to n, thus forcingg=f,,. Write f» (x) = g(x)h(x)

with h necessarily having integral coefficients because g is monic. If g(od) is not zero, then h(am) must be, and therefore w is a root of h(xP), which is congruent to h(x)P modulo p. So, g and h have a common root when reduced modulo p, contradicting the separability off,, modulo p that obtains whenever p does not divide n. This contradiction shows that W is indeed a root of g, as claimed. To summarize, we have the isomorphism

y: Ci 4 (Z/nZ)' or H ao

in the case F=Q. (ii) In the case that F=Qp, we know by Proposition 4-25 that a finite extension of F is unramified if and only if it is of the form F,,, with n relatively prime to p. Such extensions are moreover cyclic by Lemma 4-24.

(iii) Finally, we assume that F=Qp and that n=p'. We still have the injective homomorphism from G=Gal(F,,/F) into (Z/nZ)' that we constructed previously. As in part (i), to show that this map is moreover an isomorphism, it suffices to show that the order of G is again rp(n).

230


Let w be a primitive nth root of unity in F,,, so that in fact

and set

Then S" is a primitive pth root of unity. Now define X

f(x) J=O

Note thatf(x) is irreducible because g(x)=f(x+l) is an Eisenstein polynomial. [That is, the leading coefficient of g(x) does not lie in the unique prime ideal of Z., but all of the other coefficients do, and the constant term does not lie in the square of this ideal.] It follows that f (CO) =

wjP' '

_t

S"j =0

J-1

whence f is the irreducible polynomial of w over F. But of course

deg f - (p - 1)p'-' = co(n) and hence the order of G, is showing that the degree of the extension precisely q(n), as required. It remains only to show that F,/F is totally ramified. To begin, let ,r=w-1, so that ;r is a root of the irreducible polynomial g(x), which, too, has degree fi(n) over F. Then F,,=F(ir)-F[x]/(g(x)). The residual extension is still generated over F. by the image x of x. But happily g(x) =

X'P(n)

where g(x) is the reduction of g(x) modulo p, whence x = 0. (Eisenstein!) This implies that

showing that

is totally ramified, as required.

Proof of the Kronecker-Weber Theorem: The Local Case We first consider the local situation, so that F=QP. By the previous theorem, we have the following inclusions:

QP=FcF"= (P.n)=1

6.5. Abelian Extensions of Q and Qp

231

where F", F and F'b are as above, and F' is the maximal unramified extension of F in the given algebraic closure. Recall that according to our statement of Artin reciprocity, for every finite abelian extension K/F there is a canonical isomorphism between Gal(K/F) and F*/NK1F(K*). Moreover, every open subgroup of F* is a norm subgroup; that is, is of the form Nx,F.(K*) for some finite abelian extension K of F. Consequently, since F'b is the compositum (and hence direct limit) of such extensions K, we can identify

Gal(F'b/F) = lim Gal(K/F) K

with the projective completion of F*; that is, Gal(F'b/F) = lim F-*/N N

where the limit is taken over open subgroups N of P. Next recall that we have a short exact sequence

IFF**Z-*0 vp

(6.1)

which splits once we choose a uniformizing parameter rr, via yr: Z-*F*, VI(n) = 7r". For every open subgroup N of F*, this yields another split short exact sequence VP

1-+oF/oFnN-*F*/N-*Z/nZ-*0

.

With the existence of a left inverse for VF in hand, we can take the corresponding profinite limits to obtain 1- s of

m

Gal(F°b/F)-+Z

0

(6.2)

which defines a projection q' from Gal(F'b/F) onto Z. (From the proof of Theorem 1-14, we know that a profinite group is the projective limit of its quotients by open normal subgroups. Since the subgroups of n N are cofinal among the open subgroups of of , the projective limit of the corresponding quotients is precisely of itself.) And we have a final short exact sequence

1-* Gal(F`b/Fw) -). Gal(F°b/F) 4 Z -* 0

(6.3)

232


derived from the projection p from Gal(F'b/F) onto Gal(F' /F) via the natural

identification of Gal(F"IF) with Gal(F/F) _=i, where F=Fp is the residue field of F.

6-9 LEMMA- The projection q defined in sequence 6.2 may be identified with the natural projection p of Gal(F'b/F) onto Gal(F's/F). Consequently we

have an isomorphism Gal(Fb/F)= of . To prove the main statement, we must produce a compatible family of homomorphisms from the quotients v1(F*/N), N open, that appear as factors of the projective limit that constitutes the cokernel in sequence 6.2 to

the groups Gal(K/F), K unramified, that appear as factors of the projective limit that constitutes the cokernel in sequence 6.3 such that the induced map on the respective projective limits is an isomorphism a satisfying p =a -op. The following lemma contains the technical key: 6-10 LEMMA. Let K be a finite abelian extension of F with ring of integers oK. Then the following statements are equivalent: (i) The extension K/F is unramified.

(ii) NK/F(OK)= op .

Moreover, in this case F*/NK/F(K*) is a quotient of vv(F*).

PROOF. Let F'/F be the residual extension corresponding to K/F, and as usual, put f= [F': F] and n = [K: F] = ef, where e is the ramification index. Let ),rF and irK denote, respectively, the uniformizing parameters for of and 0K. Then from sequence 6.1 we get

F* = of x Z and K* = oK x Z and this is compatible with the group action from Gal(K/F). Recalling that the norm of irK is ,rf , one then easily obtains F*/NK/F(K*) = oF. INK/F( oK ) x (Z/fZ) .

6.5. Abelian Extensions of Q and QP

233

Since F*/NK,F(K*) is isomorphic to Gal(K/F), its order is n. Thus K/F is unramified if and only if n=J, which is to say, if and only if NK,F( oK )= op . The final statement is now obvious. REMARK. We have the following commutative diagram with exact rows:

1 - 1+YrKOK .j. N 1

I + rrF o,

F'*

oK

o;

1

,!, N

N

-+

-+

-+

-I

F*

1

Since the norm map on the (finite) residue fields is always surjective, it follows readily that N(oK)=oF if N(1+nrnoK)=1+frFoF. Thus if the norm fails to be surjective, it already fails at this level. PROOF OF LEMMA 6-9. Suppose that K, is a finite abelian extension of F and that K is any Galois subextension contained therein. From the diagram F*/NK,,F(K*) = OF/ INKS/F(OK1) x VF(F*)IIIVF(NK,/F(OK,))

i

y

y

F*/NK,F(K*) = OF/NK,F(OK) X VF(F9)/VF(NK/F(OK))

we see that the canonical projection on the left decomposes into the direct product of the two canonical projections indicated on the right. In the case that K is the maximal unramified subextension of K,, it follows from the previous lemma that Gal(K/F)aF*/NK1F(K*) is in fact isomorphic to the right-hand factor on the second line. Moreover, since K1/K is then totally ramified, the projection on the right is the identity map, and hence we have an isomorphism ax, : vF (F* /NK,,F (K*)) -* Ga1(K/F)

If K, is another finite abelian extension of F that contains K, with maximum unramified subextension K'/F, we have a diagram prq

vF(F*INK,,F(Kl*))

vF(F*INKi,F(Ki*))

1

1

vF(F*/NK,F(K*))

vF(F*/NK',F(K'*))

4l

Gal(K/F)

-

1

Gal(K'/F)

234


where the vertical sequences constitute aK, and aKf , respectively. To show that these maps are compatible with the projective systems is to show the commutativity of this diagram, which easily reduces to the commutativity of the lower square. But if we unwind the definitions and identifications, this follows at once from the explicit description of the Artin map given for local fields in Part One, statement (iii), of the Artin reciprocity law (Theorem 6-6). From this analysis we see that the maps aKI indeed induce an isomorphism a of projective limits

a : 1 mvF(F*/NK,,F(K*)) - lim Gal(K/F) = Gal(F'/F) where the limit on the left is taken over all finite abelian extensions of F and the limit on the right over all finite unramified extensions of F. In view of the remarks immediately following the statement of the lemma, it now suffices to show that a moreover satisfies the condition p = a o q, where p and c are defined by sequences 6.2 and 6.3 above. This amounts to checking the commutativity of the diagram

F*INK,,F(K*)

-> Gal(K1IF)

F*INK,F(K*)

-i

Gal(K/F)

where the vertical maps are the canonical projections and the horizontal maps are the isomorphisms BK1,F and OK,,. But this is no more than an element of the functoriality of the Artin map. [See Theorem 6-6, Part Two, diagram (iii).]

We now return to the proof of the Kronecker-Weber theorem. By the first lemma, which holds for arbitrary local F, we have

Gal(F"b/F') - of

.

(The compositum of local fields is local by Zorn's Lemma.) Now let F=Qp, and consider the diagram

F =F FP

Fur

F

6.5. Abelian Extensions of Q and QP

235

where FP is the extension of F obtained by adjoining all roots of unity of order a power of p. One checks easily that FP. and F' are linearly disjoint over F. (It suffices to verify this for finite totally ramified and unramified extensions; see Exercise 2 below.) Thus from Theorem 6-7, part (iii), we may deduce that

Gal(F IF-)aGal( FJ/F)=I m(Z/p"Z)" =_ Z; Accordingly, the Galois groups of both F. and Fb over F'a are isomorphic to the p-adic units, and since F'(--F.gF'b, we get an identification of F. with F'b once we prove the following:

6-11 LEMMA. Any surjective (continuous) homomorphism p:Z

Zr is an

isomorphism.

The proof of this lemma is left as an exercise. (One approach is to use that Zr is isomorphic to FP x ZP and that ZP is Noetherian as a module over Z. ) Thus we have established that every abelian extension of QP is cyclotomic.

Proof of the Kronecker-Weber Theorem: The Global Case We now consider the global case F= Q. By Artin reciprocity (Theorem 6-6),

every finite abelian extension K/Q determines a canonical open subgroup U= U(F) of CQ IQ/Q* such that CQIU identifies, via the Artin map, with Gal(K/Q). For each mz1, let U. denote the open subgroup associated with the mth cyclotomic extension F,"=Q(e2iri1'"). Since the first part of the reciprocity law implies in particular that the correspondence between open subgroups and

finite abelian extensions is bijective and inclusion-reversing, we need only show that U contains U for some m. To do this, we must first understand open subgroups of IQ and IQ/Q* somewhat better. The following result is key:

6-12 PROPOSITION. The idele group admits a decomposition as a direct product of topological groups

IQ = Q* x R; x Z"

where ix = lim(Z/nZ)" = fl ZP . Hence C. = R+ x Z" P

PROOF. Define a map :IQ_3Q* by fi(x) = sgn(x. )fl I x, IF ' P

.


236

for x=(x,,,x2,If m is a nonzero rational integer with prime factorization m=tflpja

j=1

then

4(m)=±FIlpjlpjj J=1

But the normalized absolute value of each pj with respect to itself is pi-1, and so in fact, g(m)=m. From this we deduce that fi(x)=x for all nonzero rational x; in other words, 4 provides a continuous group-theoretic section to the diagonal embedding Q*-*][Q. Thus we have Finally, it is obvious that

R; x n ZP P

whence the assertion follows.

We return to the proof that for any finite abelian extension K of F= Q, the associated open subgroup U of C. contains U. for some mz1. By the proposition above, any such U can be identified with an open subgroup of R+ x ;k'. Since the positive reals admit no nontrivial open subgroups, U must be of the

form R; x U, where the latter factor is an open subgroup in V. But an examination of the local base for the topology of k at the identity-and the Chinese remainder theorem-reveals at once that U must contain some Um, the unique subgroup of CQ corresponding to This completes the proof of the Kronecker-Weber theorem.

The Characters of CQ We conclude this section by describing all of the (continuous) characters of the idele class group of Q.

6-13 PROPOSITION. Every character of IQ that is trivial on Q* is a product of the form XI I" , where X is a character of finite order and s is a complex number.

PROOF. Let weHom,a,,(IQ,C*) with 0Q.=l. We have a topological isomorphism

6.5. Abelian Extensions of Q and Q.

237

C* - R; xS' z = re" -4 (r, a")

which we regard as an identification. Accordingly, we can decompose w as the product wrw. with w,: IQ -p R' and w,,: IQ -* S'. By the previous proposition, we may view w as a continuous character of R; x V. Since the second factor is compact and totally disconnected, its complex characters are of finite order. (See Chapter 3, Exercise 14. This is not true if, for example, we consider P-adic characters!) Thus w,j1" =1, while must be of finite order. Now put X(x)=w (p(x)), where p is the projection of IQ onto V. Since w, is trivial on both Q* and Z", it factors through the projection

IQ ->R; X I-4 1X I

and so w,(x)=i(IxI,,) for a continuous homomorphism /1: R; -* R'. Let d/i be the "differential" of 8-1 that is, di(t)=log,8(e1). Since this map is linear, it must be equivalent to multiplication by a real numbery. Exponentiating, we get wr(x) =Ixlp

By a similar argument, we see that any continuous homomorphism y:R; -). S' must be of the form a -> a" for some tER. Putting all of this together, we get w=w,wv with wr(x)=kx1X

and

for some y, to R. Hence the assertion of the proposition holds with s=y+it. REMARK. A character X on IQ that factors through CQ and has finite order must

accordingly be trivial on the R; component of the idele class group and hence is neither more nor less than a character of V. Moreover, X further factors through some component (Z/nZ)' of Z". This follows by continuity: for every rational primep,

X(Z;) = X(Z/pZ)"

238


for some nP, where we have implicitly embedded the quotient into the inverse limit in the obvious way. Hence, since X has finite image,

X(ZX)=X(f]

Zp)=X((Z/Pi'...p"'Z)' )

P

for some finite collection of primes pi. Thus the idele class character X induces a Dirichlet character, which is to say a character of (Z/nZ)" for some positive n. (Dirichlet characters are customarily extended to all of Z/nZ by assigning zero to elements not invertible modulo n.) The smallest n that affords such a factorization is called the conductor of X. Moreover, this association is patently

reversible: given any Dirichlet character, we can certainly pull it back to a character of the group Z", and hence to a unique idele class character on IQ. Thus the rational idele class characters of finite order lie in a natural bijective correspondence with the Dirichlet characters.

Exercises 1.

Let F be a global field and Pa prime of F. Show that F(P) exists and is in fact given as the compositum of all finite, unramifed extensions of K/F in a fixed algebraic closure F. [Hint: This is an exercise in cardinality. How many such K are there?]

2.

Let F be a local field with finite extensions K and L that are, respectively, totally ramified and unramified. Show that K and L are linearly disjoint over F. [Hint: Choose a basis B for L over F such that (i) BcoL and (ii) B projects onto a basis of the corresponding residue fields. What happens to a linear dependence relation over oK when reduced modulo the unique prime of the compositum of K and L? Keep in mind that the residue extension corresponding to K/F is trivial. Conclude that B remains linearly independent over K and hence that K and L are linearly disjoint over F.1

3.

Let F be a global field.

(a) Show, for every place u of F and for every positive integer n, that (F*)" is an open subgroup of F*. [Hint: First show that a subgroup G of F* of finite index is open if and only if it is closed.] (b) Show, for every place u of F and for every finite extension L of F, that the image of L* under the norm map NL,FF is an open subgroup of F*. [Hint: Note that ifn=[L:F], then (F*)" is a subgroup ofNL/F"(L*).]

Exercises

239

(c) Let K/F be a finite extension. Show, for any place u of F, that the map

fIK * Jy* 'Ju (xv) H H NK,./F, (xv )

has open image.

(d) Let K/F be a finite extension. show that NK,F(CE) is an open subgroup of CF. [Hint: First analyze the map NK,F: 4.

Let F be a local field, and let BF : CF -* CF is just F* in this local case.

FFb

be the Artin map. Recall that

(a) Show that 9F(o ,.) lies in the inertia group I = Ker (I'Fb - Gal(Fq /Fq) ) where F9 is the residue field of F.

(b) Using Part One, statement (ii), of Theorem 6-6, show that 9F induces an isomorphism of of with 1.

(c) Show that the natural topology of of is identical to that induced by the norm subgroups. 5.

Let Fbe a global field, and let OF : CF -). r;b be the Artin map.

(a) Show that if F is a number field, then 0, is surjective with kernel equal to the connected component of the identity of C,,..

(b) (Artin-Tate) Show that if F is a function field over a finite field FY, then ©F is injective with dense image. Show, moreover, that each automorphism in

the image restricts to an integral power of the Frobenius map x N xq on Fq .

SPECI,at. NOTE. It is beyond us to compose problems on class field theory and the relationship of Arlin's reciprocity law to the classical power residue symbols, prime decompositions. etc., equal to the amazing ones found in Algebraic

240


Number Theory by Cassels and Frdhlich (1968, pp. 348-364). We encourage the reader to try all of these wonderful and productive exercises.

7 Tate's Thesis and Applications

It is well known that much information on rational primes is encoded in the Riemann zeta function c'(s), which is defined by the absolutely convergent series 11

c(s) =n21I n

for complex numbers s such that Re(s)>I. Moreover, this function admits an analytic continuation to the whole s-plane, except for a simple pole at s=1, and satisfies the functional equation

f(s) _ (l - s) where b(s) =

,r-1/21-(2

)S (S)

One establishes this analytic continuation and the functional equation by making use of the Mellin transform of the theta function 9(z) = Ze2xin2x nez

and the well-known identity Ee-nn2t2 = 1-1 Ee neZ

(7.1)

neZ

for t>0. Euler was the first to study S'(s), but only for s real. He established the Euler product expansion [in fact valid in the domain Re(s)>1]

(s)=F1P (I-P 1)

242

7. Tate's Thesis and Applications

where p runs over the rational primes. He also realized that the assertion c(s) approaches infinity as s--*1' is equivalent to the infinitude of primes in Z. One could greatly generalize the zeta function with the introduction of the Dirichlet which is to say that amn aman series. Given a multiplicative sequence whenever m and n are relatively prime, one can form the series

a

L(s) _ I ns which is absolutely convergent in some right half-plane. Two very important examples are (i) L(s)=EX(n)n-, where X is a Dirichlet character (for instance, as derived from the Legendre symbol), and (ii) L(s)=EX(a)(Na) S, where a runs over the nonzero ideals of the ring of integers of a number field K and X is a character of the ideal class group CIK. (In the latter example, when X=1, the resulting series is called the Dedekind zeta function of K.) A simultaneous generalization of these two is the L-function L(s,X) associated with a (continuous) character X of the idele class group CK of any number field K. A substantial achievement of E. Hecke was to establish the analytic continuation and the functional equation of L(s,X) for any idele class character X by an enormously complicated application of generalized theta functions and the higher analogues of Eq. 7.1, which we now understand as consequences of the Poisson summation formula. One thing that Hecke's method could not describe satisfactorily was the nature of the global constant W(X), the so-called root number, appearing in the functional equation ofL(s,X). Then, circa 1950, following a sugges-

tion of his erstwhile thesis advisor E. Artin, J. Tate made use of Fourier analysis on adele groups to re-prove both the analytic continuation and the functional equation of L(s,X). In the process, Tate also established local functional equations along with a factorization of the "abelian" root number, for which he gave an explicit formula.

The basic idea of Tate was to realize the local factors and the global Lfunctions of X as the greatest common divisor of a family of zeta integrals, with a consequent generalization of Gauss sums. The key is to take a nice topological ring R such as QP, R, or A,.-and to consider integrals of the form

Z(X.(0)=5z(-x)c(x)dr

where X is a character of R' and to is a nice function on R. The functional equation reflects the Fourier duality between (X,q) and where is the Fourier transform of q, and (pi - I' )' = p l - I'- if p is a unitary character of R". Note that in the formally analogous case R=Fp, x is of order dividing (p-1), and every function q on R is a linear combination Ecwyi where y' runs over the characters of the additive group of R; that is, elements of Hom(Fp,C*). So,

7.1. Local 4-Functions

243

as suggested above, in this case, z(X 4p) becomes Ec,,B(X y.), where each g(X {v) is the Gauss sum X(a)e2arab(w)/p a=1

for some integer b(ye). When R is a local field or the adele ring of a global field, the characters yr of R are oscillatory and z(,r yr) will not converge. Here the zeta integrals make sense only for suitable functions to and may have singularities; the true analogue of the Gauss sum turns out to be the epsilon factor q) occurring in the functional equation. When R is the adele ring of a global field F, the multiplicative characters X of interest will always be trivial on F* and thus will define idele class characters.

In his thesis, Tate used some ad hoc spaces of functions over local and global fields. Here we will systematically use the spaces of Schwartz-Bruhat functions.

We end this chapter with applications, and, in particular, with a proof of the characterization of idele class characters X via their local components Xp, for p running over a set of primes of density greater than one-half.

7.1 Local c-Functions Let F be a local field with absolute value I - I and Haar measure dx. Define

d*x= c. dx

IxI

for some fixed real number c>O, which we always normalize to c=1 for F Archimedean. Then d*x is a Haar measure on F*. When F is non-Archimedean, let of denote its ring of integers, P=PF its maximal ideal, -rF the uniformizing parameter, and F. the corresponding residue field. Recall that F* is the direct product UFx SF, where OF is the subgroup of F* consisting of elements of unit absolute value and .SF is the valuation group; that is, SF= {ye R + : y=Ixl, for some xeF*}.

Then SF equals R; if F is Archimedean and qZ otherwise. (Note that OF is just the usual group of units in of in the non-Archimedean case.)

Let X(F*)=Hom.1(F*,C*) denote the space of continuous group homomorphisms from F* to C*. In this chapter, we refer to elements XeX(F*) as characters of F*. These have sometimes been called quasi-characters. Characters with codomain given as S' are here distinguished as unitary characters.

244


Hence unitary characters of F* are ordinary group characters in the sense of Chapter 3. (Admittedly, the term has been overworked.) We see that every XEX(F*) factors into the product

where p is the pullback of a unitary character on UFgF*, uniquely defined by the restriction of x and s is a complex number. This is because the compactness of U. forces its characters to be unitary, while the characters of SF are all

of the form t N t' for some se C. A straightforward calculation shows that while s may not be uniquely determined by this factorization-examine the non-Archimedean case-nonetheless, Re(s), the real part of s, always is. Accordingly, we call Re(s) the exponent of X.

The object of this section is to introduce the local L -factor L(X) associated with an arbitrary character x of F* and to realize it as the greatest common divisor of some local zeta integrals. We say that ZEX(F*) is unram fled if XI up= 1. If F is non-Archimedean, set L(X)(I-X(rrF))'

ifXisunramified otherwise.

1

If F= C, then OF is S', and X takes the form X,.,:

re'B H rsei"6

for some uniquely defined seC and neZ. (Recall that the dual group of S' is the discrete group Z; for arbitrary real n, the map e'9 H e'"a is not continuous.) We then set 114)

2

nI)

where F(s) is the traditional F-function

F(x) =1 a-`t:-l dt 0

and I'o(s)=2(2,r)-I'(s). Finally, for F=R, in which case U, =(±I), we may write X=,u I.1, with both p and s uniquely defined. Letting sgn denote the sign character x H x/IxI , we set

7.1. Local

Functions

245

jrR(S)=n ,'Zr(sl2) ifp=1 L(x)= rit (s+1) ifp=sgn. Given a character X of F* and a complex number s, the product xl - I' of course also defines a character, and one customarily writes L(s,X) for L(XI - I'). Moreover, we define the shifted dual of X by

so that

L((xl-I')")=L(1-s,x"1). Fix a nontrivial additive character yi of F; that is, a nontrivial element of F = Hom(F, S') , the ordinary dual group of (F,+). One can show that if yr' is any other additive character on F, then yr'(x) = v(ax)

for some ac=F. (See Exercise 1 below.) We will denote this character yra. It follows from this that map a H y., is an isomorphism of topological groups from

the additive group F to the dual group F, and hence we have the following result, which we shall later extend to adele groups:

7-1 PROposmoN. Any local field F viewed as an additive locally compact topological group is isomorphic to its (unitary) dual. In fact, given any nontrivial character {v of F, the mapping

F ->F a H y/a

is an isomorphism of topological groups.

O

In a case such as this of a self-dual, locally compact abelian group, we may

speak of a Haar measure dx as being self-dual if it is equal to its own dual measure in the sense defined by the Fourier inversion formula (Theorem 3-9). We will say that a complex-valued function f on F (or F*) is smooth if it is

F- for F Archimedean, and locally constant otherwise; that is, f(x)=f(xo) for all x sufficiently close to x0. In the Archimedean case, a Schwartz function f on F is a smooth function that goes to zero rapidly at infinity; more precisely, p(x)f(x) --> 0


246

as x-+oo for all polynomials p(x). A Schwartz-Bruhat function is a Schwartz function if F is Archimedean, and a smooth function with compact support in the non-Archimedean case. We let S(F) denote the space of Schwartz-Bruhat functions; this is clearly a complex vector space. Given JES(F) and the fixed additive character yr, we may define the Fourier transform off by

f(y) _ ! f(x)yr(xy) dx

.

F

Note that in this chapter it is convenient to drop the traditional conjugation of the second factor of the integrand; accordingly, this conjugation reappears in the Fourier inversion formula. While this is well-defined and in fact again lies in S(F), it nonetheless depends on the choice of the pair (yr,dx). In his thesis, Tate normalizes his measure to be self-dual relative to yi so that the identity f(x)= f(-x)

holds. We shall avoid this normalization at least for the local non-Archimedean case.

Given feS(F) and XEX(F*), we define the associated local zeta function, or local zeta integral, to be

Z(f, X) = J f(x)X(x) d*x . F'

The main result of this section is the following: 7-2 THEOREM. Let JES(F) and X = p I Is with p unitary of exponent o-=Re(s). Then the following statements hold.(i)

Z(f,X) is absolutely convergent if or is positive.

(ii) If cre(0, 1), there is a functional equation

Z(f,') = r(X,w,dr)Z(f,X) for some r(x, y,dx) independent off which in fact is meromorphic as a function of s.

(iii) There exists a factor e(X, V, dr) that lies in C* for all s and satisfies the relation Y(X, yr,dx) = s(X,1,,dx)

L(Xv) L(X)

7.1. Local 4-Functions

247

According to part (i), Z(f, X") converges for c r< 1, and so part (ii) immediately yields a mcromorphic continuation of 7_(f,X)=Z(f,u,s) to the whole splane, although initially this function is defined and holomorphic only for Re(s)>O. Moreover, from parts (i) and (iii) we deduce that

L(X)Z(.f, X") = e(X, V, dx)L(X") Z(f, X)

Since the zeta factor on the left is absolutely convergent to the left of 1 and the epsilon factor on the right is a nonzero complex number, this implies that the poles of Z(f,X) are no worse than those of L(X), which is independent off. We will see later that the "local L-factor" is given as L(X)=Z(fo,X) for some suit-

able,

PROOF. (i) Since X= ui I ` and u is unitary, we need to show that

I(f,a)= c

JIf(x)l.IxI°-' dx&, the integral reduces to

Z(f,x) =

=g-,12 r(s12)

s12

0

since in general,

F(s/2)= Je "u''2-'du. 0

Checking this against the definition of L(X), we have shown that Z(f,X)=L(X) for all characters X of this form. Next recall that

f(Y)= Je

-'e-2,azrdx=f(x).

R

(This classical formula can be proven by contour integration.) Thus we have

Z(f,X")= Jf(x)X'(x)d*x R-

which equals L(X") by what was just shown. So for X=I I', we have Y(x) =

L(X") L(X)

and we may put s(X)=s(X V/,

For F real, there still remains the possibility that X=sgn I I°. Under these circumstances take f(x)=xe'°2.

Then since sgn(x)=x/Ixl, we find that

7.1. Local

Functions

251

xe--x2 .lxl Ixl' d*x x = re- :2Ixl'+' d*x

Z(J ,X) = I

11

R'

4:L+-,) r(s + 1) 2

where the last line follows by the first computation. Thus again Z(f,X)=L(X) by definition. But contour integration also shows that

f(Y) =

iye-"y=

and so

ZU,X )=i f xe

"

=iL(X').

Thus for X=sgn I I' we have s(X)= s(x yi dr)=i.

CASE Two: F=C. We take the measure on C to be dzdi=2dxdy, which is twice the ordinary Lebesgue measure and self-dual with respect to our standard complex character W(z) = e-2"i(r+i)

Furthermore, we adjust the norm on C to agree with the module; that is, for purposes of these calculations, set

IzI=zi . As we have seen above, since C*= R; x S', every character of C* takes the form x:.,re,9

H rse"O

for some uniquely defined complex s and integral n. Put

(z)

(2rr)-1 z a 2"'i forn z 0 2"'t fore n, then Ig(a12)I2=cVol(oF,dx) [Voi(U", d*x)-q-1 Vol(U,_,, d*x)1.


256

PROOF. Write U= o' as a disjoint union of cosets modulo U,= 1+P', and note that A(a(l+,rb))=A(a)A(rc'ab)=A.(a) by definition of the conductor. Thus g(w, A) = Y A(a)w(a) j w(u) d*u U/Ur

U,

But if r0. Next we make ready for the other half of the calculation, and for this we need to compute a Fourier transform. 7-5 LEMMA. For the function f defined by Eq. 7.3, the Fourier transform off is given as Vol(P"-",dx) times the characteristic function of of for n = 0 and as Vol(Pm -", dr) times the characteristic function of P"-1 for n>0. PROOF. By definition,

f(y)= f .f(x)w(xy)dx= f w(x(y+l))dx. Pw-n

F

Let n=0. Then since the conductor of y/is P'", by orthogonality f(y) is zero if y does not he in os.. When y does lie in or., then f (y) = Vol(P-"). Now suppose that n is positive. Then if y is not in P"-1, then vv(y+l)Sn-1, and thus the product x(y+1) occurring in the integrand does not lie in P'". Accordingly, wy+i is a nontrivial character of P"-", and by orthogonality, fly) is zero. When y does lie in P"- 1, then again f (y) = Vol(P""") . O

With this fundamental technical lemma in hand, we are now prepared to compute the value of Z(f, X;") . CASE n=0. Using the last lemma and the by now familiar decomposition of integers of F into the disjoint union of subsets of a given valuation, we find that

Z(f,X;o)=Vol(P",ctx) f X;.o(y)d*y Op- (o(

= Vol(P", dx)y k20

q-k()-=) f d*y Dx

= Vol(P", dx) Vol(o,, d *x)

1

1- q-(I-S)

= Vol(P", dx) Vol(oF, d*x) L(X, o) .

258


Thus we find from this and Eq. 7.4 that

r(X,,o)=q"" Vol(P",dx) L(X:o) L(X,.o)

and

e(X,.o, W, dc) = q"" Vol(Pm, dc).

CASE n>O. Again the lemma applies, and we have at once that

Z(f,X,")= Vol(P'"-",dr) !w(u)d*u PI-I

= Vol(P',dv) 5W(-u)d*u 1+P"

= Vol(Pm ",dx)Vol(1+P",d*x)w(-1) since the conductor of w is identical to that of its conjugate. The result is a constant, as it should be, since L(X;") =1 for n positive. Accordingly, it follows from Eq. 7.5 that

qt,"-"_,Vol(P"'-",dx)Vol(1+P",d*x)w(-1) E(X,.", v,dr) = r(X,,", W,dr)=

g(w, Vs_)

Now one sees easily that

g(w,

w(-I)g(w,

and since the conductor of V*,"..,, is P", by combining the formulas above for the

epsilon factor with part (ii) of Lemma 7-4, we get the following compact formula: w dx) = I q(.-.X,-I) Here we have also used that Vol(P'"-") = q"-'" - Vol(oF.).

To conclude our analysis, we observe that in all three cases the poles of Z(J,X) are given by the zeros of the now clearly meromorphic function y(X, W, dx) = e(X, W, dx)

L(X" ) L(X)

259

7.2. The Riemann-Roch Theorem

because the regions of absolute convergence of Z(f,X) and Z(f,Xv) are, respectively, Re(s)>0 and Re(s)q,(y+S+x)=>qp(y' +x) =fi(x) yeK

y'eK

where Y'=y+S. Thus j;(S+x) = ;(x), as desired. DEFINITION. Let f be complex-valued on AK such that both f and f are normally convergent; that is, both are absolutely and uniformly convergent on compact subsets. Then we say that f is admissible.


261

7-6 LEMMA. Everyfunction feS(AK) is admissible.

PROOF. Let fES(Ax). We have to show the absolute and uniform convergence of

f over any compact subset C of A. By enlarging C, we may assume without loss of generality that C takes the form

Cmxfl R""xllo, VQS VES

where S is a finite set of finite places including those at which fl,,#l,

C. = 1I C. taken over the set S. of infinite places, and n,, is an integer for all veS. We may enlarge S to contain S and assume that fv is the characteristic function of F'," for all veS-S,,. Note that such is a compact set in the product

functions generate S(AK). Define a fractional ideal I in ox by

I= f-1 VES-S

where k,,=inf{n,,,m,,}. Suppose that f(r+z)*O for some zeC and yEK. Then y lies in P,' for all veS-S,,, and in o,, for all veS. Thus 17(z)1:5 Y, I fm(r+zm)I rer

where fm =

jlf, ES(K,)

and

Za

Vesd

But 1 is a discrete subgroup of K. (this follows, for instance, from the discreteness of K in AK), and the Schwartz-Bruhat function f has a uniform absolute

bound over the compact set C,,, with the further property that the value of decreases rapidly with z in the number field case, while f has compact, hence finite, support in the function field case. Thus for a number field, the number of r that occur in any shelf of radius B and thickness AB can grow at most as a power of B, while I f,, goes to zero faster than any polynomial; for a function field, the number of terms in the summation is finite. The

normal convergence of f follows. Since this extends at once to its Fourier transform, which also lies in S(Ax), the proof is complete.

262


7-7 THEOREM. (Poisson Summation Formula) Let fES(AK). Then f = f ; that is,

>f(r+x)=Y,f(r+x) reK

reK

for all xEAK.

PROOF. Every K-invariant function q' on AK induces a function, again denoted ip, on AK/K. For all zeK we set

(z) = Jc'(t)w(tz)dt Ar/K

where dt is the quotient measure on AK/K induced by dt on AK. This is to say that dt is characterized by the relation

Jf(t) dt = f (>f(r+t))dt = Jf(t)dt Ax/K

Ax/K reK

AK

for all continuous functions f on AK with appropriate convergence properties. (The integration variable t, as it occurs on the left and in the middle, takes values in the quotient group AK/K; nonetheless, the indicated expressions are welldefined.) We shall need two lemmas. 7-8 LEMMA. For everyfunction fin S(AK), we have

fIK=fIK PROOF. Fix zeK. By definition,

f(z)= Jf(t)w(tz)dt AK /K

J (>f(r+t))w(tz) dt . AK/K reK

Since we assume that the unitary character yv has the property w(K=1, we have that

w(tz) = w((r +t)z)


263

for all yEK. Accordingly, by definition of the quotient measure (relative to the counting measure on K) it follows that

f(z)= j (Ef(r+t)w((y+t)z) )at AK/K yeK

= j f(t) v(tz)dt AK

= f(z) and this completes the proof.

13

7-9 LEMMA. Let fES(AK). Then for every xEK, we have

f(x)= y.K

PROOF. By the previous lemma, f I K = 71K on K. Hence the summation Y f(r)w(rx) rEK

is normally convergent. In particular,

If(r)I
h(r)= Ih(r) yeK

rEX

But h(r)= j f(yx)W(ry)dy AK

=II f f(y)W(ryx-')dy

=lzlf(rx') The theorem now follows immediately.

The Riemann-Roch Theorem for Algebraic Curves When K is a function field in one variable over Fq, the previous theorem can be interpreted to yield the perhaps more familiar Riemann-Roch theorem of algebraic geometry. We shall explain this after some preliminaries.


265

A divisor on K is a formal linear combination

D=En,,v v

where the sum runs over all places v of K and each coefficient n, is an integer that is zero for almost all v. The divisors on K naturally form an additive group, denoted Div(K). The degree of a divisor D=Env is defined by

deg(D) = En,, deg(v) V

where deg(v) is the degree (over F9) of the residue field Fqy at v. Thus

q, = N(v) = gdJV)

Since deg(D+D')=deg(D)+deg(D'), we see that the degree map defines a homomorphism deg:Div(K)-+Z, the kernel of which is denoted Div°(K), the group of divisors of degree zero.

Given any feK*, we can associate a divisor, called a principal divisor, by setting

div(f) = Z v(f )v v

where v(1) of course denotes the valuation off at v. [In geometry, it is customary to write ord,(f) rather than v(f).1 Since v(f) can be nonzero only at a finite number of places, div(f) is a bona fide divisor. Moreover, it is obvious that div(fg)=div(f)+div(g). The quotient Div(K)/div(K*) is denoted Pic(K) and called the Picard group of K. Elements of Pic(K) are called divisor classes. Recall that Artin's product formula says that for all feK*,

If IA,, =F11A=1 But If Iv = qv v(f) = q--(f)des(v)

for all v, so

deg(div(f )) _ E v(f) deg(v) = 0 . Thus we see that dv(K*)cDiv°(K).

266


Now suppose that div(f)=div(g) for f and g in K*. Then div(f/g)=0, and the quotient a f/g is a unit of o for all v. From Chapter 5, Exercise 5, we know that any such a must lie in FQ*. To summarize, we have the following exact sequence of groups: div

1

Fq -* K* -a Div°(K) -* Pic°(K) -+ 0

where Pic°(K), the Picard group of degree zero, of course denotes the quotient DivO(K)/div(K*). Clearly, deg induces a homomorphism, again denoted deg, on

Pic(K), with kernel Pic°(K). Elements of Pic°(K) are called divisor classes of degree zero. We next introduce the partial ordering on Div(K) defined by

D=>n,,vzD'=Yn;v if

Vv.

V

With this, to each divisor D one may associate the following linear system of D:

L(D) _ {O}v{feK*:div(f)z-D} . Since div(f) has degree zero for feK*, we have at once that L(O)=Fq. One may further deduce from the Artin product formula that L(D)=(0) if deg(D)2g-2=deg(90, deg(x--D) 1, where

X has factorization pl - I' with p unitary, and that it defines a holomorphic function there. Define r to be X-'I - I, as in the local case. 7-16 THEOREM. (Meromorphic Continuation and Functional Equation) Z(f X) extends to a meromorphic function of s and satisfies the functional equation

Z(f, X) = Z(f, X') . The extended function Z(f,X) is in fact holomorphic everywhere except when p=I , rER, in which case it has simple poles at s=ir and s=1 +i r with corresponding residues given by

-Vol(CK1 )f(0) and Vo1(CK)f(0) respectively.

272


Here, as in Chapter 5, the symbol CK denotes the quotient IK/K*, which is the compact part of the idele class group IK/K* . The volume of C'K is taken relative to the quotient measure on CK defined by d*x and the counting measure on K*. The computation of Vol(C'K) will be done in Section 7.5.

REMARK. While we implicitly state and prove this result with respect to the standard character-which is all we shall need for the next section-it remains true for an arbitrary adele class character cv with the proviso that the associated measure remains self-dual to V. PROOF. If K is a number field, we may write, for any X of exponent greater than one,

Z(f,X)= JZ,(f,X)!dt 0

t

where

Z,(f, X) = J f (tx)X(tx) d*x 1x

(llcre the product ix takes place in a fixed infinite component of x.) For K a function field, we have

Z(f,x)= EZ,(f,x) Tao=0

with Z,(f, X) as above. We will establish a functional equation for Z,(f X) by using the Riemann-Roch theorem. To be precise, we assert the following: 7-17 PROPOSITION. The function Z,(f,X) satisfies the relation

Z,(f,X)=Z,_1(f,X')+f(0) f X"(x/t)d*x - f(0)JX(tx)d*x C,r

CX

PROOF. Since C'K is the quotient IK/K*, we have

Z,(f,X)= J (Y f(atx))X(tx)d*x = JX(tx)d*x > f(atx) C,r

ac K'

CK

oEK'

where the summation should now be regarded as the second factor of an iterated integral. We have also used the hypothesis that X=1 on K*.

7.3. The Global Functional Equation

273

To apply the Riemann-Roch theorem, we need to sum over K, not K*. This leads us to consider the following expression:

Z,(f,x)+f(o) f x(tx)d*x C'

which equals

f x(tx)d*x I f(atx). aEK

Ck

Via the Riemann-Roch theorem, we may replace the right-hand factor by

Thus

Z1(f,x)+f(o) f x(tx)d*x = f xtx) d*x

J(at 'x') aeK

Ck

Ck

= f It-'xcX(tx-')d*x EJ(at-'x) oEK

Ck

where in the second line we have replaced x by x-'. But one shows easily that this equals

Zr i(f,x")+f(O) f xv(xIt)d*x CK'

since x'=x' I.1. The formula now follows. We return to the proof of Theorem 7-16. Suppose that K is a number field. Then we may write

Z(f,x)= f Z,(J,x) f dr+ f z,(J,x)l dt.

(7.8)

0

The second integral is simply

f f(x)x(x)d*x {xE1 K : {x21 }

which converges normally for all s. Indeed, the convergence is better for small or, and since we know it converges for a>1, it must do so everywhere. But also

274


f z,(f,X)1dt=IZrXv)Idt+E

(7.9)

where the correction term is given from the previous proposition by

E= f

f(o)Xv(t_') f

I

X'(x)d*x - f(o)x(t) f X(x)d*x I I dt. C1

Cic

J

Via the substitution of t-' fort, we find that

f0 Zr (f,X") L dt = f Z:(f,Xv) I dt and hence this integral is convergent for all o by the argument above. It remains to analyze E. By the orthogonality of the characters, however, both

f X(x)d*x and f XV(x)d*x c'x

cx'

are zero if X is nontrivial on IK; hence E is likewise zero in this case. When Z=pf IJ is trivial on IK, we know that in fact X=I JS', where s'=s-ir, for some real r, and in this case,

E f [j(o)t''-' Vol(Cx)- f(0)td Vol(CK)] I dt = Vol(d

ho) - f(O) i

Since E is a rational function, we get the desired meromorphic continuation of Z(f,,X) to the whole s-plane. We have also shown incidentally that it is holomorphic everywhere if px I r' , and that when p= I. rr ' its only poles are at s=ir and s=1 +i r, with respective residues -Vol(C,r) f(0) and Vol(C,r) f (0). Finally, observe from Eqs. 7.8 and 7.9 that in fact the global zeta function may be expressed as

7.3. The Global Functional Equation

275

,

1

Z(f,x) = J Z,(.f,x) i dt+ j Z,(f,x")' dt + E(f, x)

= f f f(tx)x(tx)d*x 1 dt +f j f(tx)x"(tx)d*x 1 dt + E(f,x) 1IX

t

r

SIX

Moreover, since

f(x) = f(-x) and ((x)")" = x it follows also that

Z(f,x )= JZ,(f,xv) dt+5Z,(f,x)- dt + E(f,x") f (tx)x" (tx)d*xdt + j j f(-tx)x(tx)d*x 1 dt + E(f,x"). 1 IX

t

t

I IX

But from our explicit formula above we see at once that E is invariant under the

transformation (f, x) -4 (f, z'), and we may replace x(tx) by x(-tx) everywhere because x is an idele-class character and hence indifferent to sign. Thus for K a number field we obtain the functional equation

Z(f, Z) = Z(f, x" ) as claimed. The function field case remains. Here, by a similar argument, we get that

Z(f,x)=Z,(f,x)+I Z,. (f, z) +17-,. (f, Z') + E' >O

>o

where {t} is a set of representatives of Ix modulo IK with I

and

E' = E [f(O)Z'(t_,,) Jxv(x)d*x -f(0)x(IA) Jz(x)d*x n 1. Since log(fl I L(Xy )I) _

Elog(-

1- Xo., ( )9.'I

Re(Z E

)

Xo.V(hV)mgv )

Vm>o

m

it suffices to establish the convergence of

We will do this for the number field case, leaving the function field case for an exercise. Letting p vary over the set of positive rational primes, write

F=yy v

r

-,tea

VIr.U>O m

278


The number of v lying over any given prime p is bounded by n = [K: Q], and for each such v, the number q is a positive integral power of p. Therefore,

-.a

E5nYYp m =nlog(JJ 1-p °) p w>O But np(l-pL) is the classical Euler product, which sums to E,atn-` and is absolutely convergent if t>1.

DEFINm0N. Let x be an idele class character of Ix. Then for complex s define the Hecke L -function L(s,x) by

It is also convenient to define finite and infinite versions of the Hecke Lfunction:

L(s,xf) = JJL(s,x,) V finite

L(s, Z.) = f L(s, x.) v inrmite

The product of these two clearly gives L(s,x). Note, in particular, that when x=l, we have

L(s,If) _ JJ

1

where P is the prime associated with the finite place v and N is the absolute norm map. In particular, if K= Q, we obtain the Riemann zeta function S(s)

P1-p'

in (Re(s)>1). For arbitrary K, L(s, lf) is called the Dedekind zeta function of K, and denoted SK(s). Just as in the rational case, for Re(s) >I we have 1

N(a),

.;K (s) = e

where a runs over the set of nonzero ideals of ox.

7.4. Hecke L-Functions

279

REMARKS. (i) Hecke actually considered a slightly different definition involving a generalized ideal class character; we explore this subsequently in Exercise 13.

(ii) What we here denote L(s, Z) is sometimes written as A(s,x); the notation classically denotes that which we have called L(s,X.). 7-19 THEOREM. Let x be a unitary idele class character. Then L(s,X), which is

a priori defined and holomorphic in (Re(s)>1), admits a meromorphic continuation to the whole s -plane, and satisfies the functional equation L(1- s, XV) = e(s, X)L(s, X)

where

e(s,x)=fle(xJ I') EC*

.

Moreover, this meromorphic continuation is entire unless X= ieR, in which case there exist poles at s = i z and s =1 +1 r with respective residues - Vol(C') and IN(4)1-112 Vol(CK). PROOF. First we claim that the asserted functional equation of L(s,X) will follow once we show that it is meromorphic everywhere. Indeed, if we choose a factorizablef=®,,fES(AK), we will have (by Section 5.1) that with respect to any adele class character on Ax,

z(hx")=rizcl ,X: ) Specializing to the standard character W. and appealing to the global functional equation of Section 7.3 and the local one of Section 7.1, we obtain

r, E(s,Xr)L(1-s,Xv) = E(s,X)L(1-s,Xv) V

L(s, Z,)

L(s, X)

Hence we shall have our functional equation and meromorphic continuation, provided that L(s,X) is indeed meromorphic. That L(s,X) is meromorphic in turn follows at once from Theorem 7-16 if we can establish the existence of a function f=®,, j,ES(AK) with the property that

Z(.fx1 ' I') = h(s,x)L(s,X)

(7.10)

for a nonzero meromorphic function h. But we can see now that for every place v, we have already, in the proof of Theorem 7-2, constructed a local function such that

280


where h, is entire and everywhere nonzero for all v and, in fact, equal to I for almost all v. Indeed, in the real and complex cases the given Schwartz-Bruhat Z(./v,Xvl. I`)=L(s,X) precisely. In the p-adic case, the local functions f, yield standard characters are given by yp(tr(x))

where v lies above the rational prime p. If we again let m denote the exponent of the conductor of y/v, which is zero for almost all v, then our previous construction takes the form

jv(x)

if x e P""-"

=0

otherwise

where nv is the exponent of the conductor of X,,. From our prior calculations (see Eqs. 7-4 and 7-5) and the normalization of d*x,, to give volume q- on the local unit groups, it follows that q-m,.(=-vz)L(s,

Z(ff,XVI.1")

Xv)

if n =0 )L(s,Xv) otherwise.

9

Moreover, since nv is zero for almost all v, we see that jv is the characteristic function of ov for almost all v, and thus f=®vff, does indeed define a function f in S(AK) such that Z(.JXl I3) has the requisite property of Eq. 7.10.

Finally, for X=1- I-" the expressions given for the residues are derived as follows from the corresponding residue formulas of Theorem 7-16. Locally we have everywhere that f,(0)=1, which establishes the residue at s=0. To com-

pute the residue at s=1, we can, via the global functional equation, simply compute the residue at s=0 for the Fourier transform of fv. But in this case, nv 0 for all v, so by construction fv is the characteristic function of P'", because this is precisely the conductor of y/ . It then follows from Exercise 7 below that

jv(0) =I Thus taking the product of these over all v completes the proof.

0

7.5. The Volume of CK and the Regulator

281

7.5 The Volume of CK and the Regulator Let K again be a global field. Since the residue of OK(s) at s=1 involves the volume of CK = IK /K*, it is our next order of business to compute it. First recall the following definitions and results from Chapter 5. For any finite set S of places of K, let us now define the S-ideles of K by

VS)

IK.S

with norm-one version given by IiK.s

= IK r-1 'K,$

Observe that here we do not require that S contain the infinite places of K, and thus this is a slight, but compatible, extension of the definition given previously in Chapter 5. Note also that IK.O = IK m is compact. We shall prepare ourselves for the eventual volume calculation with three preliminary steps. STEP ONE. Assume henceforth that S is nonempty. Then we have the short exact sequence

1+IKS'K*/K*=Ixs/K*nIKS+Cx=IK/K*_+Cxs=IK/K*'IK

*1

where CK s has finite order, say, hs. (We proved this in Section 5.3 only for the

case that S contains S,,, but the present extension is trivial.) Consequently,

Vol(C'' ) = hs Vol(IK s /K*n IK,s)

(7.11)

and our calculation is reduced to finding the volume of the second factor.

STEP Two. Assume henceforth that K is a number field. Take S=S,, the set of Archimedean places of K, and write Card(S) as the sum rl + r2, where rl is the number of real embeddings into a fixed algebraic closure of Q, and r2 is the number of nonconjugate complex embeddings. Define the logarithmic map as follows:

A : Ir.s< -+ Rs° = R'j" (X) H (log IXV1y )VESm

282


This is clearly a continuous homomorphism. Next define a hyperplane H in

R"' by the equation

it,, +2Et,,=0 v real

v complex

7-20 LEMMA. The logarithmic map has the following properties: (i)

Im(A) = H.

(ii) Ker(A) = Iir.ra(= IK.e)

PROOF. That the image of A lies in H follows from two facts. First, that fl I Xvly =1 vES,,

Vx = (xv) E 1I .S.

and second, that the normalized absolute value I.1v coincides with the usual absolute value for v real and the square of the usual absolute value for v complex. Moreover, given t=(tv)EH, we can consider the idele with x, 1 for v finite and xveKK for of ordinary absolute value tv. Then by construction A(x)=t, and this proves part (i). Since lxjv=1 for all v for any xEIKO, it is obvious that Ix OcKer(A). Now suppose conversely that x E IK

belongs to Ker(A).

Then loglxjv 0, i.e., l xjv 1, for all veS,,,. Thus X E IK.O(= IK,e) . The restriction of A to K* n IK sq is called the regulator map and denoted reg(x). Recall that in fact, oK = K*n IK.sb

From the lemma above, we see that

Ker(reg) = IK., n K* = pK where pK is the set of roots of unity in K. Put wK = Card(pK)

and L = reg(oK) .

Then L is a discrete subgroup of H, which in turn is isomorphic to R' where r=rl+r2 1. Also, since IK/K* is compact, HIL is likewise compact. Thus L is a full lattice in H.

7.5. The Volume of C'K and the Regulator

283

STEP THREE. As our final preliminary for the calculation at hand, note that clearly, I1 .. admits the product decomposition

v complex

v reel

flu,

v finite

where U denotes the subset of elements of K,, of absolute value one. We can thus establish a Haar measure on IK H given by the product measure whose factors are defined as follows:

For v real, this is the counting measure on U,=(±I). For v complex, this is the ordinary Lebesgue measure on S. For v finite, this is the normalized measure d*x defined previously. Thus 2

2fr

N(4)-''2

for v real for v complex for v finite

where D,, is the different for finite v. Since one knows from Appendix B, Section 2, that the absolute value of the discriminant dK (see also Chapter 4, Exercises 13 and 14) is given by

I dKI = f N(A) finite

we get, relative to this measure, Vol(II 0) = 2'1(2n)'=I dK I-12

(7.12)

We may now combine the results of our three preliminary steps to obtain the following marvelous formula: 7-21 THEOREM. Let K be a number field. Then we have - Res1 I SK (s) = Vol(CK) =

2'1(2,r)' hK RK WK 1dKI

where hK is the class number of K and RK is the regulator of K; that is, the volume of HIL relative to the quotient measure induced by the map .7* defined below.

284


PROOF. From step two we have at once the commutative diagram below, all of whose rows and columns are exact: 0

1

1

4, reg 1

-3

L

og

UK

- 0

I

4,

W

A 1

3

I

1

IK.m

IK.s

--3

H

-*

0

4, 1

-> IK.glfUK

-3

HIL - 0 4,

0

The formula now follows at once from Eqs. 11 and 12. Note that RK is computed

with respect to the quotient measure induced (ultimately) by the measure on I;r.e established in step three and our standard measure on the idele group. REMARK. Let T (=PK) denote the torsion subgroup of oK , and define L' by

L' _ (oxIT) ® Z which we regard as a free Z-module. Define a homomorphism

reg': L' -F R'1"2

(u,m) N reg(u)+mJ

is the vector with 8,=1 (respectively, 5,=2) for v real where (respectively, for v complex). Then it follows from the remarks above that reg' is an embedding whose image is a full lattice in R'1+'2. This induces an isomorphism

(reg'®R):L'®RR" One can check that RK is none other than the absolute value of the determinant of this map relative to integral bases drawn from L' on the left and Z'1"2 on the right. There are many situations in arithmetic and algebraic geometry where two lattices like this are sitting in a Euclidean space and the determinant of an

7.5. The Volume of CK and the Regulator

285

associated map from one to the other, as above, gives special values (or residues) of more general zeta functions. This is an active area of research. We will conclude this section with the function-field case. Here we fix any place v° of K and take S= (v°). As previously, let Div°(K) denote the group of divisors of degree zero on K; that is, finite formal sums D of the form

D=ln,,v

(n,,eZ)

V

0. (See Section 7.2.) Again we have the Picard with deg(D) group of degree zero given by the quotient Pic°(K) = Div°(K)/K*

Recall that in forming where each fEK* defines the divisor this quotient, we are using implicitly that Z. v(J)deg(v)=0, which is to say that f has as many zeros as poles, when counted with multiplicities. Indeed, this is true of any element x of adelic norm one because by definition, II,,I x,, IV 1. We may once again extend the divisor map from K* to a larger structure:

div:IK S -*Div°(K)

x H >v(x,,)v . V

Since x is a local unit at almost all v, we know that v(x,,) is zero almost everywhere, and so the formal sum on the right has at most finitely many nonzero components. Moreover, arguing as above, E,,v(J)deg(v)=0 because xe IK.S, and thus the extended map is indeed well-defined. Finally, suppose that xeKer(div). Then v(x,,)=0 for all v, and therefore xEIKO n,, U,. From the equality IK 0 n K* = F4 , we get the short exact sequence

1->IK0/Fq*--* IAS/K*->Pic°(K)--*O Thus Pic°(K) is compact and discrete, and therefore finite. With these considerations in mind, we have the following function-field-theoretic version of our previous theorem: 7-22 THEOREM. Let K be a function field over F9. Then

-Res,, 4K (s) = Vol(CK) =

1

1089

Cand(Pic°(K))

.

286


7.6 Dirichlet's Class Number Formula In this section, we shall specialize our results to the base field Q, prove a factorization formula, and then recover the class number formula of Dirichlet for cyclotomic fields.

Recall from Section 6.5 that there is a natural identification of idele class characters x of Q of finite order and the Dirichlet characters %D. By abuse of notation, we will write x to denote either. 7-23 PROPosrrION. Fix mz 1, and consider F.= Q(e2iriIm). Then we have

(s) _ fl L(s, x)

(7.13)

where the product runs over all the Dirichlet characters x of conductor mx dividing m.

PROOF. It suffices to show that for each rational prime p the corresponding local factors are the same. In other words, we have to show that for t=p-°,

fl(1-tf)=fl(1-x(p)t)

(7.14)

x

VIP

where jV-- [Fm QP]. Since Fm is Galois over Q, jv f is the same for each of the g places v lying above p, and likewise, the corresponding ramification indices have a common value, which we denote e. We know further that

q(m) = [Fm: Q] = erg Hence the left-hand side of Eq. 7.14 may be rewritten and factored as follows: F1

t'

(I - tf )g = 11 (1-zt)8

VIP

:1=1

Accordingly, one may obtain Eq. 7.14 at once from the following lemma, the proof of which we leave as an exercise.

7-24 LEMMA. For every fth root of unity z there are g characters x (mod m) such that x(p)=z. Note that on the right side of Eq. 7.13, the factor corresponding to the trivial character is precisely the Riemann zeta function, which has a simple pole of

7.6. Dirichlet's Class Number Formula 287

residue I at s=1. Thus by the residue computation of 4.M(s) at s=1, we have the following formula: (21r}p(m)rz

W.

h,,,R, = fl L(l, X) .

(7.15)

,mode ,rsl

l dml

Here the symbols wm, dm, hm, and Rm denote, respectively, the number of roots of unity, discriminant, class number, and regulator associated with Fm.

One can do better. Suppose that K is any finite abelian extension of Q. By the Kronecker-Weber theorem, K is a subfield of some Fm. Put G=Gal(K/Q),

which is a quotient group of G.=Gal(Fm/Q)-(7JmZ)'. Then its Pontryagin dual G is a subgroup of Gm. Dirichlet characters X modulo m are naturally identifiable with elements of Gm. A refinement of the proof of Eq. 7.13 gives in this case the factorization

x(s)= HL(s,X) XEQ

In particular, 6x(s) is a factor of 4FT(s). Now using the residue formula for 4K(s) at s=1, we obtain the following powerful theorem.

7-25 THEOREM. (Class Number Formula) Let K be a finite abelian extension of

Q with Galois group G, number of roots of unity wx, class number hx, regulator Rx, and discriminant dx. Let r,(K) and r2(K) denote, respectively, the number of real and nonconjugate complex embeddings of K into an algebraic closure of Q. Then we have 2^(x)(2,r)' WK ,fl dx

hxRx = n L(l,X). X:t

One may wonder at the importance of this formula. The reason is that L(1,X) admits a concrete expression. Indeed, as we shall see in Exercise 14 below, by elementary Fourier analysis one may obtain an explicit formula:

7-26 PROPOSITION. Fix m z 1, and let X be a Dirichlet character modulo m. Then

L(l, Z) _

-g(X) jX(a)log(1-ezxia/m) M

amodm

288


where g(X) is the Gauss sum Y-X(a)ez r;a/m a mod m

More explicitly, we have

g(x) 2 L(l,x)=

tri

m

if x(-1) _ -1

X(a)a

amodm

0

-g(x) m

YX(a)logjl-e2,r;a/ml

if X(-1)=+1

amodm

When this proposition is combined with the preceding theorem and then specialized to quadratic fields, we get the following beautiful result of Dirichlet.

7-27 THEOREM. (Dirichlet) Let K=Q(h) be a quadratic field of discriminant ,52 =D, and let XD be the quadratic Dirichlet character associated to K by class field theory; that is, for all rational primes p not dividing D, p splits in K if and only if XD(p)= 1. Then the root number for XD is given by

W(XD)= g(XD)(-1)' E(-1,+1)

where r=0 (respectively, 1) i[(-1)= 1 (respectively, -1). Moreover:

(i) IfD0), and U is the full group of units in oF. Show directly that IG12=q". Then appeal to Eq. 7.6 to deduce (once again) that IW(w)j=1.

10. (Tate) Let F be a non-Archimedean local field with uniformizing parameter if, and let w be a unitary character of F* with conductor ,r"o,,. Let a be an ideal of of such that a2 divides 7r"oF. Put b=a-';r"oFcoF. (a) Show that there exists an element cEF such that coF= x"DF and also

w(1+t) = yi(c-'t)

for all tEb. [Hint: Suppose that a#oF, so that it divides a. If t,zeb, then tze,r"oF. Consequently, w(1 +t) w(I +z)= w(I +t+z), and therefore the map that sends t to w(1+t) is a character of the additive group b, and this extends to one on F. Now appeal to the isomorphism of F with its dual.] (b)

For c as in the previous part and F now assumed to have characteristic zero, show that W(w) =

N(ba-1)-1/2

Y w(C-'x)

y/(c-'x)

x (I+a)/(1+b)

NOTE. When a=o', this result is identical to Eq. 7.6. (c)

For w unramified, show that W(w)= w(ir)d if ,rd of 7)

.

(d) Let E be a number field containing W(w). Then show that for every place v of E not dividing the residual characteristic p of F, we have IW(w)lp=1. Using the Artin product formula, conclude that if E has a unique place u above p, then also I

(e) Suppose that n"oF=;ra. Then, using part (b), show that z=(w(c-I)W(w))2 lies in the cyclotomic extension E=Q(e2'"/p') for some rz1. Show further, using part (d), that z must be a root of unity.

Exercises

303

(f) (Lamprecht, Dwork) Suppose that w is either unramifred (that is, n=0) or wildly ramified (that is, n22). Show that W(w) is a root of unity. [Hint:

This is clear if n is even. When n is odd, write ;rôF=ira and apply the previous part.]

11. Let K be a global field. For any idele class character X=(X,), put

W(x)=rjW(xo (a)

(b)

.

Show that W(X) W(x-')=1. In particular, if X is unitary, then I W(X)l =1.

Let K be a number field, and let X be quadratic or trivial. Conclude from part (a) that W(X)=±1. [This number is called the sign of the functional equation of L(s,X).] Show, moreover, that if X is unramified everywhere, then W(X) = X (DK)

where j is the associated character of the class group C/K (cf. Proposition 5-19). [Hint: Use the previous problem.] (c) Let X be a quadratic idele class character, and let E/K be the quadratic extension corresponding to the open subgroup Ker(X) of CK by class field theory (Theorem 6-6). Note that i&(s)= CK(s) L(s,X) and deduce the formula

s(s,X)=

(ldKl/Idel'i2)1-2,

where dK and dF are the discriminants of K and E, respectively.

(d) (Hecke's Theorem, Serre's Proof) Let K be a number field. Then prove: THEOREM. (Hecke) The ideal class of aK is a square in the class group CIK.

[Hint: Observe that it suffices to show that j (7.) =1 for every quadratic character x of CIK. Then appeal to parts (b) and (c) of this problem.] 12. We consider here the Fourier transforms of Schwartz-Bruhat functions.

(a) Let F be a local field. Show that for every fnS(F), its Fourier transform likewise lies in S(F). [Hint: For F Archimedean, this is a well-known classical fact. In the non-Archimedean case, use that f is a linear combination of

304


characteristic functions of the basic compact sets P", where P is the unique prime associated with F.) (b)

Let K be a global field and assume that fES(AK). Show that the Fourier transform of f is likewise in S(AK). [Hint: First prove that S(AK) is generfor all v, ated by factorizable functions ®J,,, where each f, lies in and then use the previous part to establish the result in this special case.]

(c) Use the results of Chapter 3 to show that the Fourier transform map S(AK) -+S(AK) extends to an isometry LZ(AK)->LZ(AK).

13. (Hecke Characters) Let K be a number field of degree d, let S be a finite set

of places containing S., and let JK(S) be the group of fractional ideals prime to S. Furthermore, let IK(S) denote the subgroup of IK consisting of ideles y=(y,,) such that yv 1 at every place v in S. (a) Define a: 'K(S) -+JK(S) by

a(y) = TT Pv(r,. ) vials

Show that a is a surjective homomorphism with kernel U(S) = (yeIK(S):y,,e ov' , Vv finite) . Conclude that every character X of IK that is unramified outside S defines a character X of JK(S). (b) Let /7 be a Grossencharakter (or Hecke character) of K, which is to say a

homomorphism from JK(S) to C* for which there exists an integral ideal M with support S, complex numbers s,,...,s, and integers m,,...,md such that for every aEK*(M), one has d

Ma)) _ fJ a. (a)' l aa(a)I

r-

where {c.} is the set of embeddings of K in C. Show that everyQ is of the form ,j for some character X:IK-sC* that is trivial on K*. [Hint: Show that fl defines a character of the ray class group CIK(M) and then lift it to the idele class group CK.]

Exercises

305

(c) Let 40 =X for an idele class character X unramified outside S. Show that CD is in fact a Grossencharakter. [Hint: Take

M = 1-1 PVv Ves

where n is the exponent of the conductor of y,. For aeK*(M), show that a(a) by using first that la,J= l for all v in S and second that X is trivial on K*.[

14. Fix an integer kz 1, and consider the polylogarithm function 'k(Z)

z" n

k

Note that 1,(z)=-log(1-z). The special case of 12(z) is called Euler's dilogarithm function.

(a) Show that the series is normally convergent in the open unit disk {fzf1 (respectively, k=1).

(b) Let X be a Dirichlet character of conductor m;-> I. For w an mth root of unity, put

G(X, w) = J X(c)-c c mod m

Let L(s,X) be the Dirichlet L-series, and assume that X is not the trivial character if k=1. Put X(-1)=(-ly. Then show that

L(k,,)=? m

L

G(X,e 2,ibi",)[lk(e2,.iba,)+(-1),lk(e-2'' "))

l

15b5m/2

[Hint: Expand X in terms of the basis (e2xib/" :05b<m) of C[Z/mZ].]

(c) Show that G(X

e-2,ribim)

= X(b)G(X,

e-2,riim) .

Simplify part (b) accordingly for k=1 and deduce Proposition 7-26.

306


(d) For any kZ1, show that there is an elementary expression for the Dirichlet series L(k, X) if X(-1) _ (-1)k.

15. Let xbe a Dirichlet character of conductor m;-> 1. As above, put

Show that W(X) =

GW (-i)r m

where r=0 (respectively, 1) if x(-1)= I (respectively, -1), and

G(X) _ I

X(a)e2aia/m

amodm

16. Let K be a number field and let S be a set of primes in oK. One says that S has lower Dirichlet density

a .!

I

1

¢(S) =1im inf log

S i Pes

I N(P)s

1

provided that the indicated Jim inf exists. The upper Dirichlet density 8(S) is likewise defined via the lim sup.

(a) Show that every S has both a lower and an upper Dirichlet density. [Hint: Use Proposition 7-31, part (ii), and also that a bounded sequence in R has an infimum and a supremum.]

(b) Show that S has a Dirichlet density (as defined previously) if and only if

8(s) = s(s) . (c) Let T be any finite set of primes in oK. Show that 8(SvT) =¢(S) and 8(S v T) =5(S) . Conclude, in particular, that any finite set of primes has density zero.

(d) Let . ' denote the set of degree-one primes of K; that is, prime ideals of oK

such that N(P)=1. Show that for any set S of primes, 5(S-)= 8(S) and 8(S n J,0') = 8(S). Conclude that the set of primes in oK of degree greater than one has density zero.

Exercises

307

17. Let E/K be a finite abelian extension of number fields with Galois group G=Gal(E/K). Let U= NK,K(CE). Recall that CK/U is isomorphic to G via class field theory. (See Theorem 6-6.) For any character X of G, let j denote the corresponding character of CK that is trivial on U. Keeping in mind that the extension E/K is abelian, for any prime P of K, we let

(E/K) gyp =

P)

denote the corresponding Frobenius element. Now set

r(P) =

(X(c,) if P is unramified in E or if X is trivial otherwise

0

and put L(s, X) _ I) (1- X(P)N(P)-' )_t P

(a) Show that L(s,X) converges absolutely in (Re(s)>1) and further admits a meromorphic continuation to the whole s-plane, with no poles except possibly a simple one at s=1, which occurs only when X=1. Show, moreover, that there is a functional equation relating s to 1-s. [Hint: Use Theorems 6-6 and 7-16.1

(b) Show that 5s(s)= LTL(s,X) %EG

[Hint: Write the left-hand side as an Euler product over the primes P of K.1 (c) Show that 1 lim x~I' log

s-1

X (P) PE9K N(P)'

=

fI

if =1

0

otherwise.

The following four problems lead the reader through a proof of a version of the Tchebotarev density theorem, here reformulated in terms of Dirichlet density. Recall that.9'K f denotes the set of finite places of K, or, equivalently, the set of prime ideals of 0K.

308


THEOREM. Let E/K be a finite Galois extension of number fields with Galois group G, and let C be a conjugacy class in G. Set

SK (C)Pelf: (EP )

C}.

Then 8(SK(C)) exists and equals Card(C)/Card(G). We shall also need the degree-one version of SK(C), which we define as S'K(C) = SK(C)n-IK" .

The first problem is well known and deals with the abelian case. The next three encapsulate the ideas from an elegant proof due to 0. Schreier.

18. Let E/K be a finite abelian extension, so that C reduces to a singleton set {a} for some ac=G. We recall from Chapter 3, Exercise 13, the discrete form of the Fourier inversion formula: if f is a complex-valued function on G, then

f(g)= Card(G)xE f(x)x(g) where

I(x) = I f(g)x(g) aeG

(a) Given aeG, define fa:G-+C by

fa(g)=

X(a-'g) rEO

Show that fa is in fact the characteristic function of (a).

(b) For all geG show that

I PEsK(C)

l= E I f.(g) N(P).,

N(P)

PESK(C)

1 - x(P) N(P),

= Cani(G) rECT P

(c) Show that 8(SK(C)) exists and equals 1/Card(G)=Card(C)/Card(G). [Hint: Use the previous exercise.]

Exercises

309

19. Suppose that E/K is any finite normal extension of number fields. Let G be the associated Galois group, and let C be a conjugacy class in G. Suppose

further that E' is an auxiliary finite abelian extension of K with Galois group G' such that E and E' are linearly disjoint over K. Put L=EE'. (a) Show that LIK is normal with Galois group -9=GxG'.

(b) Choose any conjugacy class C'={a} in G', and let H be the subgroup generated by CC' in G. Put F=LX, and consider the compositum FE'. We have the following diagram:

L=EE' HI

F

F

E

E'

K

Show that FE' is a normal extension of F with Galois group given by

Gal(FE'/F) = H/Hr-(C) - (C') c G'. Using the previous problem, conclude that

S({QED': FQF

C'})=

where f'= Card((C')).

(c) For P ESx(C), define aa(F)=Card({ Q E_IP,'(C) : QIP}). Show that

ap(F)=Cu(ff) Card({rES:rCr-' EH}) [Hint: Use Proposition 6-2.]

(d) Put f-- Card(C). Assuming that f I f', show for P E S'K (C) that

I-i nH=O LIK P

20. We continue in the context of the previous problem and always assume that

flf'.

310


(a) Show that lim

s_l+

aP(F)

-log(-) f s-1 PESK(C.C') N(P)s ~ J 1

1

'

where

SK(C,C')={PeSK(C):(EPK =C'}. (b) Show that

s(Sx(C,C'))= f nn' where n and n' denote Card(G) and Card(G'), respectively. [Hint: Show that aa(F)=nn'1ff'.]

(c) Let

nJ =Card({v' EG':fIo(a')}). Show that 8(S,r (C)) Z

fn nn

where again 8 denotes lower Dirichlet density, as defined above. [Hint: Any absolutely convergent summation indexed over P e SK (C) can be rewritten as a double summation, first over Q'EG' and then over

(PESK(C): E /K =d}. Use this and appeal to part (b).1

21. We continue in the context of the previous two problems.

(a) Show that we can choose G' such that the quotient n f/n is arbitrarily close to 1. [Hint: Suppose that f has prime factorization r

a

f = fj P;J J=I

Then take G' of order

(a1 > 0)

Exercises

311

such that for each j, pilJ- piJ-' elements of G' have order a power of p".] (b) Use the previous exercise and the first part of this one to show that

s(SK(C))z

f n

(c) Show that

d(SK(G-C))z

n-f n

[flint: G-C is the disjoint union of conjugacy classes different from C.)

(d) Show that 6(SK(C)) exists and is given by f/n=Card(C)/Card(G). [Hint: Use part (c) to conclude that S(SK (C)) :!g

f n

But the lower and upper Dirichlet densities always exist, and the latter bounds the former from above.]

22. Let K be a function field in one variable over a finite field Fq. The first parts of this exercise lead to a proof of the following result:

THEOREM. The functionCK(s) is a rational function of T=q-1. More precisely, there is a polynomial P of degree 2g, where g is the genus of K, with P(0)=1, such that

P(T) OK(s)

(1-T)(I-qT)

Furthermore, P(1)=h, the order of the divisor class group Pic°(K). (See Section 7.5.)

REMARK. A celebrated theorem of A. Weil asserts in addition that P(T) lies in Z[t] and admits a factorization over C of the form

312

7. Tate's Thesis and Applications 2j

fl(1-ajT)

rl

with each a. an algebraic integer of absolute value q"2. Consequently, every zero

of SK(s) lies on the line (Re(s)=1/2)-in other words, the Riemann hypothesis holds for K! But a proof of this result lies far beyond the scope of this book.

(a) Show that there exist infinitely many places v of K such that N(v)=q. Hcrc the norm of v simply measures the cardinality of the residue field associated with the corresponding local field K,,. [Hint: Suppose not. Then

C (s)=fl(1-q-'). v6S

(1-q fs)-t Ves

for a finite set S, with f, 2 for all v outside S. Deduce from this that in fact, log ;,Ks) has a finite limit as s--> V, and thus derive a contradiction to Theorem 7-19.] (b) Show that OK(s) is a meromorphic function of q,*. [Hint: Show that the image of I.1KR : IK R; is qZ, and then use that S"x (s) = L (1.1' ) ]

(c) Using part (b) and Theorem 7-19, show that 4K(s) takes the form given in the theorem above, with P(q-°) an entire function.

(d) Using the functional equation forCK(s), deduce the following functional equation for the numerator:

P(T)=gST22P( T). 9

Deduce that P must be a polynomial, and (by using Theorem 7-22) that P(1)=h. (e) Let X be any idele class character of K. Show that L(X) is a rational function of T=q ". [Hint: Reduce to the unitary case and argue as above. Provided that X* I - I'

,

the function L(X) is in fact a polynomial in T. ]

23. Let K be a number field. For x> 0, put 7rK(x) = Card((P prime in oK:N(P)Sx))

.

The "prime number theorem" for K then asserts that as x-ioo,

Exercises

YrK (x) -

x logx

313

(7.21)

For K= Q, this was conjectured by Riemann and proven independently by Hadamard and de la Vallee-Poussin in 1896. The object of this exercise is to deduce this theorem from the following theorem for Dirichlet series: THEOREM. (Tauberian Theorem for Dirichlet Series) Let

D(s)

natty,

be a Dirichlet series with nonnegative coefficients an that satisfies the following conditions: (i) D(s) is normally convergent in {Re(s) > 1 }.

(ii) D(s) is invertible on the line {Re(s)=1 ) except for a simple pole at the points= I of residue a. Then

Ya -- ax nsx

asymptotically in x.

(a) Show that S'K(s)l4x(s) = D(s)+ip(s)

for some function p(s) holomorphic and nonzero in {Re(s)t 1 } and D(s) the Dirichlet series defined by an=logn, if n is the norm of some prime P, and an=0, otherwise. (b) Using the results of Section 7.7, show that the function D(s) of the previous part satisfies the hypotheses of the Tauberian theorem and has residue 1 at s=1. Deduce that

g(x) = E logN(P) - x N(P)Sx

(c) Put bn=an/logn. Show that for any mZ1, we have

314

7. Tate's Thesis and Applications )TK (m) =

I

n

_ 'g(m)

al

m

og n

+ h(m)

where

_ I h(m)_ n<mg(n)log(n+1) 1

1

Ilogn

[Hint: an=g(n)-g(n-1).J (d) Using that g(n)-n, prove Eq. 7.21 by showing that h(m) is o(m/logm); that is,

lint

m-sao

logm m

h(m) = 0 .

[Hint: Show first that

(logn

for alln>1.]

log(n+1)

nlog2n

Appendices

The two appendices address, respectively, the elementary theory of normed linear spaces and the factorization properties of Dedekind domains. In both areas, our goal is to review fundamental definitions and results that have been used frequently in the main exposition; hence the discussions below are sharply limited. We shall indicate comprehensive sources in the references.

Appendix A: Normed Linear Spaces In the main, this appendix addresses common topological constructs on normed linear spaces and particular aspects of LP-spaces and LP-duality. We show first that the topological possibilities for a finite-dimensional normed linear space X are essentially limited to one, and then discuss for general X the weak topology and the weak-star topology on the continuous dual X*. The discussion culminates in Alaoglu's theorem. The final section defines the LP-spaces for locally

compact Hausdorff spaces and states without proof the duality theorem for spaces belonging to conjugate exponents.

A.1 Finite-Dimensional Normed Linear Spaces If X is any normed linear space (real or complex), we let S' (X) denote the set of

elements of X of norm 1. Similarly, B'(X) denotes the set of all elements of norm less than or equal to 1, the so-called unit ball.

Recall that 1,(C") is the complex normed linear space whose underlying vector space is C" with norm given by II(aa)II = laJ + .. + la"I

for (al)EC". [We shall also write a for (a') when convenient.] The point of this brief discussion is to show that every complex normed linear space of dimension n is isomorphic to li(C") in the category of complex normed linear spaces with morphisms given as continuous linear maps. It follows from this that any two finite-dimensional complex normed linear spaces of the same dimension

316

Appendix A: Nonned Linear Spaces

are likewise isomorphic. Note that this discussion applies equally well to the category of real normed linear spaces. We begin with some easy preliminaries. Let X be a complex normed linear space of dimension nz 1. Choose a basis x...... x" for X, via which we may define the following map, which at the least is an isomorphism of vector spaces:

4p:l(C")->X

i Evidently, II ta((a,))IIs supllx,ll

I aiI

whence 97 is bounded with respect to the 1,-norm and hence continuous. We wish to show more, namely that g' is an isomorphism of normed linear spaces. The key technical point lies in the following lemma: A-1. LEMMA. There exists a positive constant esuch that for all elements (as.)E 11(C") of unit norm, Ilq,((ai))IlZ c

PROOF. As (ai) ranges over the compact set S'(11(C")), II4p((ai))II at some point

assumes a minimum value. This minimum cannot be zero, since the vector space isomorphism qp has trivial kernel, and the zero vector is manifestly not of unit norm. Hence the minimum is indeed a positive number e, as claimed.

We deduce from the lemma that c is moreover a topological isomorphism as follows. Let xeB'(X) be nonzero and suppose that g-1(x)=a. Then. I1q,(a/Ilal1)IIZ_-

and so by construction,

Il ea'(x)Ii=llallsllxll 1 Hence qrI is likewise bounded, and therefore continuous. Thus we have proven a fundamental result: A-2. THEOREM. Let X be a finite-dimensional complex normed linear space of dimension n. ThenX is isomorphic to II(C"). Consequently, any two finite-

dimensional normed linear spaces of the same dimension over C are isomorphic.

317

A.2. The Weak Topology

As a further consequence of the isomorphism q we may observe that every finite-dimensional complex normed linear space is moreover complete, and therefore a complex Banach space.

A.2 The Weak Topology Let X be a (real or complex) normed linear space. Then the given norm defines a topology on X via the associated metric. This is called the norm, or strong,

topology, and with respect to it, X is of course a locally convex topological vector space. We shall now introduce a second natural topology on X, comparable to the norm topology, but still of a somewhat different character.

DEFINITION. Let X* be the continuous dual of X. The weak topology on X is defined to be the coarsest topology such that each map x*nX* is continuous. Since the inverse image of any open neighborhood of 0 must be weakly open for each x*, the weak topology has a neighborhood base at 0 given by sets of the form

N(0;x*,...,x,!;e)={xEX:)x*(x)I<e,j=1,...,n).

(Al) .

We may deduce from this that the weak topology is Hausdorff and satisfies the first axiom of countability; hence with regard to convergence we may deal with sequences rather than nets. Thus it follows at once from the definition that a sequence (xj) in X converges weakly (i.e., converges in the weak topology) to a point x0 in X if and only if for each x*EX*, we have x*(xo) = lim x*(x,.) . J__

(A.2)

These observations yield the following fundamental result: A-3. PROPOSITION. Let X be as above. Then the following assertions hold:

(1) The weak topology is indeed weaker than the norm topology.

(ii) X is also a locally convex topological vector space with respect to the weak topology.

PROOF. (i) Recall that in an arbitrary first countable topological space, the closure of a subset Y is exactly the set of points that can be obtained as the limits of convergent sequences in Y. Clearly, Eq. 2 implies that strong convergence implies weak convergence (since each x* is continuous), and so if a subset of X is

318

Appendix A: Nonmed Linear Spaces

weakly closed, it is also norm closed. Hence the weak topology is indeed weaker than the norm topology. (ii) Let {(x1,y1)} be a weakly convergent sequence in XxX with limit (xo,Y0). Then the sequences {x} and {y,} converge weakly to x0 and yo, respectively. Now since each x* is additive, the definition of weak continuity and the ordinary continuity of addition on X with respect to the norm topology yield the weak limit

{x+y1} -+ xo+yo . Hence we have the following commutative square of mappings and weak limits: (x , Y1) I

F--)

xi + Y.i I

(xo>Yo) H xo+Yo showing that addition is weakly continuous on XxX. A similar argument shows that scalar multiplication is likewise weakly continuous, and hence X is at least a topological group with respect to the weak topology. It remains to show that X is weakly locally convex. For this, we consider Eq. A. 1. Suppose that

x,yeN(O;x*,...,x,*;e)

.

Then for all indices j and real numbers t, 05t51, the triangle inequality yields

Ixj*(tx+(1-t)Y)I s tlx*(x)I+(l-t)Ixj*(Y)l s e whence tx+(1- t)y e N(O;x*, ..., x,*; r). HenceX is locally convex.

The weak topology on a normed linear space X also gives rise to a weak dual: the space of all linear maps from X to the ground field that are continuous

with respect to the weak topology on X. The notation for this construction might have proved too much of a challenge (perhaps X,*.k), but fortunately, the weak dual coincides with the ordinary norm dual.

A-4. PROPOSITION. Let X be a normed linear space. Then the weak dual ofX coincides with X*, the ordinary norm dual of X.

PROOF. Let W be an arbitrary space with topologies r and r' and suppose that r is weaker than r'. Then for any fixed topological space Y, every map WAY that

A.3. The Weak-Star Topology

319

is r-continuous is automatically r'-continuous: if the inverse image of every open subset of Y is open in the r topology, it is certainly also open in the r' topology. In particular, since the weak topology is weaker than the norm topology, every weakly continuous linear map from X to C is a norm continuous linear map from X to C. Conversely, by definition of the weak topology every norm continuous functional is also weakly continuous.

We show next that in a Banach space, weakly compact sets are norm closed

and norm bounded. The key to this is the uniform boundedness principle, which we now recall: Let X be a Banach space, and let Y be a normed linear space. Suppose that 9 is a subset of Hom(X, Y), the space of bounded operators from X to Y, such that for every xoX, the set {T(x): Te 9) is bounded in Y. Then 9 is a bounded subset of Hom(X, Y). A-5. PROPOSITION. Let K be a weakly compact subset of the Banach space X Then K is norm closed and norm bounded. PROOF. Since K is weakly compact, it is weakly closed, and hence norm closed.

It remains to show that K is bounded. Since each element of the dual space is weakly continuous, for each element x*EX, we know that x*(K) is a compact, hence bounded, subset of C. But consider the natural isometric embedding of K into X** defined by k(x*)=x*(k) for each keK. (The isometry, hence injectivity, follows from the Hahn-Banach extension theorem.) Under this identification, K is a subset of Hom(X*, C) which is bounded at each point of its domain. Hence by the uniform boundedness principle, K is norm bounded in X**, which is to say norm bounded in X.

A.3 The Weak-Star Topology Let X be a (real or complex) normed linear space, with norm continuous dual X*. Then every element xeX gives rise to an evaluation map VJEX** in the usual way: v,,(x*) = x*(x)

for all x*eX*. Indeed, since X is locally convex, the Hahn-Banach separation theorem asserts in particular that the mapping x H vx is an embedding. Accordingly, we shall often simply write x(x*) for vx(x*).

DEFINITION. The weak-star topology on X* is the coarsest topology such that each evaluation map vx for xeX is continuous.

320

Appendix A: Normed Linear Spaces

Note that in the special case that X=X**, the weak-star and weak topologies on X* coincide by definition. Thus they can differ only to the extent that the evaluation maps from X do not exhaust X**.

We can also characterize the weak-star topology on X by limits and neighborhoods. A net {x2*} in X* converges in the weak-star topology to a point xo* in X* if and only if for each xeX we have xa*(x) = Jim xt*(x)

.

(A.3)

Moreover, since the inverse image under an evaluation map yr of a neighborhood of zero in C must be weak-star open for each xeX, the weak-star topology has a neighborhood base at 0 given by sets of the form

N*(0;xl,...,x,,;e)=(x*eX:Ix*(x,)I<e,j=l,...,n).

(A.4)

Thus as above, we have that the weak-star topology is both Hausdorff and first countable. These facts suffice to prove the following result: A-6. PROPOSITION. LetX be as above. Then the following assertions hold: (i)

The weak-star topology ofX* is weaker than the weak topology.

(ii) X* is a locally convex topological vector space with respect to the weak-star topology. PROOF. (i) Rewriting Eq. 4 explicitly in terms of evaluation maps, we have N*(0; xl,..., x,,; e) = (x* E X: I vz, (x*)I < e, j = 1, ..., n) .

Thus we see that weak-star neighborhoods are in fact weak neighborhoods, and this proves (i).

(ii) The argument that X* is a locally convex topological vector space with respect to the weak-star topology is entirely similar to the corresponding argument for the weak topology. C]

We shall next show that the weak-star dual of X* is precisely X. Note that this argument is not as shallow a formality as the proof that the weak and norm duals of X are identical. In fact, the key is the following purely linear algebraic result:

A-7. LEMMA. Let V be a vector space over the field k, and let f, gj, ...,g be elements of the dual space. Suppose further that Ker(f) LD (n Ker(gj)). Then f lies in the span ofg...... g,,.

A.3. The Weak-Star Topology

321

PROOF. By induction on n. If n=1, we have Ker(f)QKer(g,), so that both fand

g, factor through the induced maps f,g, : V/Ker(g,)-*k. Since the quotient space has dimension no greater then one, f is certainly a scalar multiple of and hence f is likewise a scalar multiple of g1. Now assume that n>1. Let fl,911,

denote the respective restrictions of the indicated maps to

Then clearly,

Ker(f+) Q m-' n Ker(gi! ) whence by induction, f j is a linear combination of the maps g1 1, ...,g,r1 1. Thus for some family of scalars A,_., A.,,_, ek, the map

fAjgi J

vanishes on Ker(g ). But then Ker(f)

Ker(g ), and so by the case n =1 it

follows that I is a scalar multiple of g,,. Hence f is indeed a linear combination of the g,, as claimed. A-8. PROPOSITION. Let X be a normed linear space. Then the weak-star dual of X* coincides with X itself. PROOF. Let

x as an element of the dual of X* via the evaluation map. Then by definition of the weak-star topology, x is a weakstar continuous map on X*, and so only the converse is interesting. Let f be a weak-star continuous linear map from X* into C. Then

U= {x*eX* : I f(x*)+ I define LP(X)cL(X) as follows:

L°(X) = (fEL(X):IfI°eL'(X)) . One defines a norm II II, on LP(X) by the formula UP

HAP=If If1Pdx By virtue of Minkowski's inequality, II I1, is indeed a norm on L '(X), and in

fact, LP(X) constitutes a Banach space with respect to this norm. One extends these considerations to the case p=oo as follows. Assume that f : X- (0,00] is a measurable function. Consider the subset S of R defined by

S= {aeR: p(f-'(a,oo]) = 0)

.

Now define s, the essential supremum off by the formula

324

Appendix A: Normed Linear Spaces

s=

JnnfS ifS*O 00

otherwise.

Note that if S is nonempty, then indeed sES by virtue of the formula

f-'(Is,-ll =Uf-' ((s + n

and the fact that a countable union of measurable sets is measurable. Given any feL(X), define Il!IIL to be the essential supremum of If I. Accordingly, set L`°(X) _

is finite} .

Elements of L °°(X) are called essentially bounded functions. The general inclusion

(f+g)-'((a + b,ol) c f-'((a,xl) Li g-'((b,cel) shows at once that II II,, is indeed a norm on L °°(X). In fact, L "°(X) is a Banach space with respect to this norm.

Duality We conclude this synoptic review with a key duality statement for L"-spaces, but first we must introduce a technical restriction on our locally compact Hausdorf space X. DEFINITION. A topological space X is called -compact if X is the countable union of compact subsets.

Clearly the metric spaces R" and C" are a compact, since each is the union of balls of integral radius. Moreover, one can show that every locally compact, compact Hausdorfspace is normal. We now state the main result: A-10. THEOREM. Let X be a locally compact, a--compact Hausdorfspace with Radon measure p. Let p and q satisfy the relation

where 1:5p, q :!g co and l /co is defined to be zero. Then for each pair of functions ffL"(X) and geL4(X),

A.4. A Review ofLP-Spaces and Duality

325

(fIg>= Jfkdu X

is finite. Moreover, the mapping

L°(X) --s L4(X)*

fH(fl-) defines an isometric isomorphism from L p(X) to (L 9(X))*.

Note that this theorem clearly extends to the case that X is the disjoint union of cr-compact sets, a condition met by every locally compact topological group G, as demonstrated by the following argument:

Let K be a compact neighborhood of the identity of G. Then K admits a symmetric subset V, also a neighborhood of the identity, which we may assume is closed and hence compact. The subset V in turn generates a subgroup H of G, which is manifestly the countable union of compact sets:

H=U(V) J=1

.

k=1

Finally, G is the disjoint union of cosets of H, thus proving our claim.

326

Appendix B: Dedekind Domains

Appendix B: Dedekind Domains We now survey the elementary theory of Dedekind domains. In particular, we demonstrate the crucial property that the group of fractional ideals of a Dedekind domain is free on its prime ideals, and we examine the behavior of primes under extension. Throughout, all of our rings are commutative with unity.

B.1 Basic Properties Let A be an integral domain, which is to say that {0} is a prime ideal of A. Then a nonempty subset ScA* is called multiplicative if it is closed under multiplication. In this case, we can construct the ring As = {als : aEA, sES}

via the usual quotient construction: a/s=a'/s' if and only if as'=a's. This is called the localization ofA at S. Given any sES, we clearly have an embedding of A in As defined by sending aEA to as/sEAs, and one sees that we may assume that 1 ES.

In the particular case that S=A*, the localization As is the full fraction field of A. This example has an important generalization. Let P be any prime ideal of A. Then S=A -P is a multiplicative set. (We accept the convention that a prime ideal is a proper ideal.) In this case, we write A, for A. and speak of the localization of A at P. Localization at an arbitrary multiplicative set S has the following key property with respect to prime ideals. B-1 PROPOSITION. Let A be an integral domain and let S be a multiplicative subset ofA. Then the maps

Q' i-4Q=Q'nA Q ,-4 Q' = QA

constitute a mutually inverse pair of order preserving bijections from the set of prime ideals ofA5 to the set of prime ideals ofA that have empty intersection with S. PROOF. Exercise.

0

A ring having exactly one maximal ideal is called a local ring. We see at once that in such a ring, the complement of this unique maximal ideal consists precisely of the group of units. It follows from the proposition that if P is a

B.1. Basic Properties

327

prime ideal of A, then AP has only one maximal ideal, namely PAP. Therefore the localization of a ring at a prime ideal is always local, and if Q is any ideal of A not contained in the prime P, then clearly QAP blows up to AP. Thus localization at a prime ideal vastly simplifies the multiplicative structure of a ring.

REMARK. Some algebraists prefer to reserve the term local ring for a Noether-

ian ring with a unique maximal ideal; they then call what we have defined above a quasi-local ring.

The ideals of an integral domain and the full ring itself are determined by localization at maximal ideals in the following sense. B-2 PROPOSITION. Let A be an integral domain. Then (i)

We have that

A=nAM M

where the intersection is taken over all maximal ideals ofA.

(ii) Let JocJ, be a chain of ideals in A such that JoAM=J,AM for all maximal ideals M ofA. Then J,=J,. PROOF. Let x=y/z (y,zEA) lie in the given intersection, and assume that xeAM for every maximal ideal M. Consider the set 1= {aeA : aynAz}, which is clearly an ideal of A. If M is any maximal ideal, then y/z=y'Iz' for some ring elements

y' and z' with z' not in M. Hence z'el, and thus I does not lie in any maximal ideal. Accordingly, I=A, and so 1 EI, whence yeAz and x=y/zEA. This proves part (i). The proof of part (ii) is similar, but in this case we show that for every xeJ,, the ideal is all ofA, showing at once thatxeJo. 0 Local rings admit a special case that will be of utmost importance to us. A principal ideal domain having exactly one nonzero prime (hence maximal) ideal is called a discrete valuation ring. Note that this definition excludes fields. If the unique prime ideal of A is generated by the irreducible element n; then it is called a uniformizing parameter for A, and it is unique up to a factor in A'. One sees at once that every nonzero element of A factors as uir" for some unit u and unique nonnegative integer n; moreover, every ideal of A has the form Air", again for a unique n. This brings us to the key definition of this appendix. DEFINITION. Let A be a Noetherian integral domain. Then A is called a Dedekind domain if for every nonzero prime ideal P of A, the localization AP is a discrete valuation ring.

328


Our most interesting examples will arise shortly in connection with integral elements. For the present, we make the following elementary observations, all of which follow immediately from the first proposition above: (i) Any principal ideal domain is a Dedekind domain.

(ii) Every prime ideal of a Dedekind domain is maximal. (iii) The localization of a Dedekind domain at any multiplicative set is likewise a Dedekind domain.

Our goal for this section is to demonstrate some fundamental equivalent characterizations of a Dedekind domain, but before doing so, we must review the notion of an integral element over a ring.

Integral Elements Let B be an extension of the ring A, so that inclusion is a unitat ring homomor-

phism. Then an element xEB is said to be integral over A if there exists a monic polynomial p(t)EA[t] such that p(x)=0. B-3 PROPOSITION. Let A, B, and x be as above. Then the following four statements are equivalent: (i)

The element x is integral over A.

(ii) The ring A [x] cB is finitely generated as a module over A.

(iii) The ring A[x] is contained in a subring A' of B that is finitely generated as a module over A. (iv) There exists an A [x]-module L, finitely generated over A, such that the only element ofA that annihilates L is zero.

PROOF. That (i) implies (ii) follows at once from the observation that if x satisfies a monic polynomial of degree n in A[t], then by Euclidean division A[x] is generated by 1,x, ...,x" as a module over A. Clearly, (ii) implies (iii), and (iii) implies (iv). (For the latter, take L=A[x] itself, which contains 1.) Thus it remains only to show that (iv) implies (i). Let L be as stated, and let b....., b, be a set of generators for L over A. Then for each index i we have an equation xb, = Y.?.ybj J-,

for some

A. Now consider the rxr matrix M defined by

B. I. Basic Properties

329

M = (x8;1 - 2; )

where S,, is the Kronecker delta. Then we have the matrix equation

Mb=O where b is the column vector whose components are b,,...,b,. Multiplying both sides of this equation by the adjoint of M shows that d=det(M) annihilates every b, and hence all of L. By hypothesis we then must have that d=0, and therefore x satisfies the monic polynomial det(tS,t-2,1) inA[t]. Thus x is integral l] over A, and this completes the proof. If x,yeB are both integral over A, then A[x,y] is finitely generated as an A. module, and hence by part (iii) above, their sum, difference, and product are likewise integral over A. Thus we have the following immediate corollary:

B-4 COROLLARY. Let A and B be as above. Then the set of all elements of B O that are integral over A is a subring of B containing A. The ring consisting of all elements of B integral over A is called the integral closure of A in B. One deduces easily from the proposition above that the operation of taking the integral closure within a fixed extension is idempotent. If A is equal to its integral closure in an extension B, we say that A is integrally closed in B. We say that an integral domain is integrally closed (without reference to an extension) if it is integrally closed in its fraction field. We leave it to the reader to show that every unique factorization domain is integrally closed.

Characterization of Dedekind Domains With the notion of integral closure in hand, we can now state our main result on the characterization of Dedekind domains. B-5 THEOREM. Let A be an integral domain, and assume that A is not a field. Then the following three statements are equivalent:

(i) A is a Dedekind domain.

(ii) For each maximal ideal M of A, the localization A. is a discrete valuation ring, and each nonzero element of A is contained in onlyfinitely many prime ideals.

(iii) A is Noetherian, integrally closed, and every nonzero prime ideal is maximal.

330


Before proving this, we need to do a little more elementary commutative algebra. B-6 PROPOSITION. Let A be a Noetherian ring. Then (i) Every ideal ofA contains a product of prime ideals.

(ii) There exist distinct prime ideals PI,...,Pr and corresponding positive integers mi, ...,mr such that

{0}=fP'"-'

.

1=1

Assume further thatA has zero divisors and that every nonzero prime ideal ofA is maximal. Let the prime ideals P1, ...,Pr and integers ml, ...,mr be as above. Then we have, moreover, that

(iii) A f j A/P!"f . (iv) The prime ideals P,'..., P, are the only prime ideals ofA. PROOF. Part (1) follows by Noetherian induction: If I is maximal among ideals not containing a product of primes, then there exist x and y in A such that neither x nor y lies in I, but the product xy does. Then (Ax+I)(Ay+1)cI, and both

factors, by virtue of being strictly larger than 1, contain products of primes; hence so does I-a contradiction. Part (ii) is an immediate corollary. Part (iii) then holds in consequence of the Chinese remainder theorem, once we note that if I and J are distinct maximal ideals of A, then I' and J" remain comaximal for all positive m and n. Finally, part (iv) follows at once from part (iii). O B-7 COROLLARY. Let A be a Noetherian ring for which every nonzero prime ideal is maximal. Then every nonzero ideal of A is contained in only fi-

nitely many prime ideals. In particular, every nonzero element of A is contained in only finitely many prime ideals.

PROOF. Let I be a nonzero ideal of A. If I is prime, it is contained in only one proper ideal, namely I itself. Otherwise, Al! is not an integral domain, and we can apply parts (ii) through (iv) of the preceding proposition to this quotient. O We may now proceed to the proof of the main theorem. PROOF OF THEOREM. (i) =:> (ii). If A is a Dedekind domain, it certainly satisfies

the condition of the preceding corollary, and hence assertion (ii) clearly holds

B.1. Basic Properties

331

for A. (Note that {0} is not a maximal ideal, whence the localization statement.)

(ii)=(iii). First, since A is the intersection of its localizations at maximal ideals, each of which is integrally closed by virtue of being a unique factorization domain, we have thatA itself is integrally closed. Second, ifPcQ is any proper chain consisting of a prime ideal P and maximal ideal Q, then PAQcQAQ is likewise a proper chain of prime ideals in the discrete valuation ring A.-an impossibility. Hence every prime ideal of A is maximal, and it only remains to show that A is Noetherian. Let 1 be any nonzero ideal in A and let XEI, with x itself nonzero. Then there exist only finitely many prime ideals P...... P, of A that contain I. At each cor-

responding localization-which is a principal ideal domain-we have IA pi _ yi APi

for some y,,...,y,, each of which, as one shows easily, we may assume also to lie in 1. Now consider the ideal

which is clearly contained in I. On the one hand, if P is any prime ideal of A not containing x, then JAP and IAP both blow up to A. On the other hand, if P is any prime ideal of A that does contain x, then P=P1 for some j, and by construction,

y1ApgJA,gJAP=yjAP and again JAP=IAP. Thus Jc! constitutes a chain of ideals that collapses locally at every prime ideal, and therefore =1 by Proposition B-2. Accordingly, our original ideal I is finitely generated, whence A is Noetherian.

(iii)=(i). Since A is given as Noetherian, we need only show that for each nonzero prime ideal P, the localization AP is a discrete valuation ring. We know already that AP has a unique prime ideal, because the nonzero primes of A are maximal by hypothesis. Since this localization is also integrally closed and Noetherian, the proof of the characterization theorem is complete if we can show that any Noetherian, integrally closed domain having precisely one nonzero prime ideal is also a principal ideal domain. Let B be such a ring, with Q its unique nonzero prime.

Given xeB-B", consider the nontrivial quotient B/Bx as a module over B. For each nonzero element y+Bx in this quotient, let I(y)cB denote its annihilator. Then because B is assumed Noetherian, among these annihilators there is a maximal element, I,,, corresponding to, say, e (B/Bx)*. One checks readily that I is prime and hence equal to Q. So while y itself does

332


not lie in Bx. Thus z =ya,/x does not lie in B, and therefore z cannot be integral over B. But certainly QzcB, whence Qz is an ideal of B. We claim that in fact, Qz=B. Assume contrariwise that QzcQ. Then we have: Qz is a B[z]-module. Qz is finitely generated over B.

Since B is an integral domain, the final statement of Proposition B-3 yields a contradiction: namely, that z is integral over B. Thus Qz=B, and setting -r=z-1, we have Q=B?r (and in fact, -ir will be our uniformizing parameter). We now complete the proof that B is a discrete valuation ring. Let I be any nonzero ideal of B. Then 1=Iz-lzclz, whence we have an ascending chain of B-modules

which must become stationary. But if Iz"=Iz' , then Iz" is a B[z]-module, finitely generated over B, and again we have the contradiction that z is integral over B. Hence only a finite part of this chain can remain in B. So assume that for some nonnegative integer n,

Iz"cB but Ii"" cr B. Then it cannot be the case that Iz"c;.-Q=Brr, or else multiplying by z=rr' yields another contradiction, and therefore Iz"=B. This to say that I=Bjr". Hence B is principal, and the proof is complete.

Factorization of Ideals We shall now develop another critical property of a Dedekind domain. The point is that while a Dedekind domain need not exhibit unique factorization at the level of elements, it does so nonetheless at the level of ideals.

B-8 PRoposmoN. Let A be a Dedekind domain. Then every nonzero ideal I of A has a unique factorization into a product of prime ideals. In fact, the

ideals appearing in this factorization are precisely those prime ideals containing 1.

PROOF. Clearly we may assume that I is a proper ideal. Then I is at least contained in some prime ideal, and Corollary B-7 implies that there are only finitely many such primes. Let these be Pj,...,Pr. They are also precisely the primes of A/I, and we know from Proposition B-6 that I;2 r 11 PIRJ J-1

B.I. Basic Properties

333

for some positive integers ni. By the Chinese remainder theorem, r

r

B=A/f Pj"J =f AIPj-J i=1

i=1

and by localization it is easy to see that every ideal in the factor rings appearing on the right takes the form

pJ.'/P"i for some nonnegative mi. Hence the image of I in B takes the form r

F1 Pj J / Pi"f j=1

and therefore

We see at this point that the mi are in fact positive, or else we would contradict the hypothesis that I is contained in each of the Pi. This shows the existence of the asserted prime factorization. We can easily deduce uniqueness, again by localization. Suppose that r l=I

=TTQ; 1=l

for a second family of prime ideals (Qi). Then localizing at any P. yields, on the left-hand side, AP 2rJmJ J

where ii is the uniformizing parameter for the local ring APB. Hence each Pi must correspond to a Qi, and corresponding factors must manifest the same exponent. This completes the proof. O REMARK. A strong form of the converse of this proposition holds: an integral

domain in which every ideal is the product of prime ideals is necessarily a Dedekind domain.

334


Fractional Ideals and the Ideal Class Group Let A be a Dedekind domain with field of fractions K. A fractional ideal of K is a nonzero finitely generated A -submodule of K. Let JK denote the set of all such

fractional ideals, the so-called ideal group of K. This clearly contains all the ideals of A and, in particular, A itself.

Given I,JEJK, define the product IJ as for ordinary ideals; this clearly remains in JK. Moreover, for IeJK, define I-'={xeK: IxcA}. One checks easily that I-1 EJK, but perhaps it is not obvious that I"tI=A. For ordinary ideals this follows from Proposition B-8 by localization; for arbitrary elements of JK we need an extension of the cited proposition. This is given by the first part of the following theorem.

B-9 THEOREM. Let A and K be as above. Then every element of IeJK has a unique factorization of the form

I=HPM' i=1

where the exponents m, may be positive or negative. Consequently, JK is a free abelian group on the prime ideals of A.

PROOF. We can construct an element xEA such that IxcA. Noting that I=(Ix)(Ax)-', we may then apply Proposition B-8 to both Ix and Ax to get our factorization. The rest is a straightforward exercise.

Elements of the form Ax in JK, with xeK* are called principal fractional ideals, and these constitute a subgroup denoted P. The quotient CIK=JK/PK is the familiar ideal class group of K. Note well that for a general Dedekind domain, ClK need not be finite. This shows that one essentially needs some analysis to supplement the abstract algebra in Chapter 5.

B.2 Extensions of Dedekind Domains In this section we give some further fundamental definitions and state, without proof, some key results that arise in connection with extensions of Dedekind domains. Indeed, the theorems that appear here in essence define our interest in this rich class of rings. Throughout, let A be a Dedekind domain with fraction field K, and let LIK be a finite extension of K. Then the integral closure of A in L is a subring of L, which we denote B. Clearly, AQB. The following theorem is paramount; it has the immediate particular consequence that for a global field K, the ring of integers uK is a Dedekind domain.

B.2. Extensions of Dedekind Domains

335

B-10 THEOREM. Let the ring extension B/A and the field extension LIK be given as above. Then B is also a Dedekind domain. While we omit the proof, we will remark that one approach to proving this result is to advance in two steps by introducing an intermediate field E such that one story of the resulting tower is separable, while the other is purely inseparable. In the separable case, we may use the nondegeneracy of the trace map to show that the corresponding integral closure is Noetherian, and then proceed to show that it satisfies part (iii) of our characterization of Dedekind domains (Theorem B-5). In the purely inseparable case, we may use part (ii) of our characterization: the argument is reducible by localization to the case that A

is a discrete valuation ring, and one then shows that B is likewise a discrete valuation ring. From this theorem and the results of the preceding section, we have at once that given any prime ideal P of A, 9

PB =11

j

where Ql,...,Q2 are the prime ideals of B that lie above (that is, contain) P. Hence in this entirely algebraic setting we now become reacquainted with two old friends from Chapter 4. DEFINITIONS. The number ej defined above is called the ramification index of Qj over A. The number

j = (B/Qj: A/P) (that is, the degree of the extension of residue fields) is called the residual degree of Qj over A.

Of course, every prime Q of B lies over some prime P of A, namely QrnA. Thus Q is said to be ramified over A if Q has ramification index greater than one or if the corresponding extension of residue fields fails to be separable. Otherwise, we say that Q is unramifred over A. In the same vein, a prime P of A is ramified in B if it is divisible by a prime of B ramified over A; otherwise, P is unramified.

We have these familiar-looking results that relate ramification index to residual degree:

B-11 THEOREM. Let P and Ql,...,Q8 be as above. Then the following statements hold:

336


(i)

The summation Eel j equals the dimension of B/PB over the residue field A IP and, moreover, is bounded by the degree of L over K.

(ii) IfL/K is separable, then in fact, 2ejj=(L:K). (iii) If LIK is Galois, then all of the ei have a common value e and all of the j have a common value f, moreover, efg= (L: K).

The Norm of an Ideal Again A is a Dedekind domain with fraction field K, LIK is a finite extension of K, and B is the integral closure of A in L. Recall that for any xEL, NL,K(x) and trL,K(x)

the norm and the trace of x, are, respectively, the determinant and the trace of the K-linear endomorphism of L that sends to xy. One knows that

NL,K(x)=(fla(x))[LXL

and

trL,K (x) = [L:K]r y-a(x)

where both sum and product are taken over a full set of embeddings of L over K into a fixed algebraic closure of K, and [L : K], is the inseparable degree of the extension. It follows at once from these formulas that both maps send elements ofB into A.

We may extend the norm map NL,K:L-*K to ideals. If I is an ideal of B, define NL,K(I) to be the ideal of A generated by all of the images NL,K(x), where

x ranges over I. In the special case that K=Q, the ideal NL,K(I) is contained in Z and therefore is generated by a unique positive integer, which we shall de-

note simply N(I). This is called the absolute norm of an integral ideal in a number field. We summarize the most important properties of the norm and absolute norm for ideals in the two following propositions. B- 12 PROPOSmTioN. Let the extensions B/A and LIK be given as above and as-

sume further that the latter is separable. Suppose that I is a nonzero ideal of B with prime factorization

I = fIQ7J

.

J=1

Put P.=Q1nA and set f, equal to the residual degree of Qf over A. Then P

NL,K(I)=F1 J1r1 i=1

B.2. Extensions of Dedekind Domains

337

B- 13 PROPOSITION. Let K be a number field and let A be the integral closure of Z in K. Then for any nonzero ideal I of A, we have N(I) = Card(A/I).

The Different and the Discriminant The extensions A/B and LIK remain as above, with the continuing assumption that L/K is separable. If J is any subset of L, then J' denotes its dual subset, which is defined by

J'= (xeL:tr,K(xJ)cA) . One can show that if J is a fractional ideal of L, then J' is likewise a fractional ideal of L. The dual subset corresponding to B itself is called the inverse different of B/A. The different of the extension, denoted DB,A, is then the inverse fractional ideal of the inverse different. Since DB,A is the inverse of a fractional ideal that contains 1, it is in fact an ideal of B, and one can show that this ideal is determined locally. The following general relation shows how the different is fundamental to the calculation of the dual of any fractional ideal J of L:

J'=

(Da1A)-1J-1 .

Moreover, one has this essential connection with ramification:

B-14 THEOREM. Let Q be a prime ideal of B. Then Q is ramified if and only if Q divides the different DB,A. In fact, Q'-1 divides DB,A, where e is the ramification index of Q over A. REMARK. As an immediate corollary, we have that only finitely many primes of B are ramified over A.

We now develop one further ring invariant. Let x1,..., ...... x,, be a basis for L over K. Then A(x,,...,xn) = det(trL,K(x;xj))15ijSn

lies in K and is called the discriminant of the basis x,, ...,xn. If each x, lies in B,

then A(x,,...,x,,) lies in A. Thus as xi,...,xn range over all bases of L over K that are contained in B, the elements generate an ideal of A, denoted A(B/A) and called the discriminant ideal. This again may be determined locally, and the discriminant gives us a criterion for ramification at the lower level:

B-15 THEOREM. Let P be a prime ideal ofA. Then P ramifies in B if and only if P contains the discriminant ideal A(B/A).

338


Finally, we state the relation between the different and the discriminant; this is mediated by the norm:

B-16 THEOREM. Let the rings A and B and the separable extension L/K be as above. Then we have that NLiK(Dau) = A(B/A) .

That is, the discriminant is the norm of the different.

The reader should refer to the exercises from Chapter 4 for a development of the different and the discriminant for the integers of local and global fields.

References

Artin, Emil. The Gamma Function. (Translated by Michael Butler.) New York: Holt, Rinehart and Winston, 1964.

Artin, Emil. Algebraic Numbers and Functions. New York: Gordon and Breach, 1967.

Aupetit, Bernard. A Primer on Spectral Theory. New York: Springer-Verlag, 1991.

Bruhat, F. Lectures on Lie Groups and Representations of Locally Compact Groups. Bombay: Tata Institute of Fundamental Research, 1968.

Cartan, H. and R. Godement. Thdorie de la dualite et analyse harmonique dans

les groupes abeliens localement compacts. Ann. Sci. Ecole Norm. Sup., 64(3), 79-99, 1947.

Cassels, J. W.S. and A. Frdhlich, eds. Algebraic Number Theory. New York: Academic Press, 1968.

Dikranjan, Dikran N., Ivan R Prodanov, and Luchezar N. Stoyanov. Topological Groups: Characters, Dualities, and Minimal Group Topologies. New York: Marcel Dekker, Inc., 1990. Folland, Gerald B. Real Analysis: Modern Techniques and Their Applications. New York: John Wiley and Sons, 1984. L.J. Goldstein. Analytic Number Theory. New Jersey: Prentice-Hall, 1971.

Gorenstein, Daniel. Finite Groups. New York: Harper and Row, 1968. Hall, M. The Theory of Groups. New York: MacMillan, 1959.

Hecke, E. Mathematische Werke. Gtttingen: Vandenhoeck & Ruprecht, 1959. Janusz, Gerald J. Algebraic Number Fields. New York: Academic Press 1973.

Ireland, Kenneth and Michael Rosen. A Classical Introduction to Modern Number Theory (Second Edition). New York: Springer-Verlag, 1990.

Kaplansky, Irving. Commutative Rings (Revised Edition). Chicago: The University of Chicago Press, 1974.

340

References

Koblitz, Neal. p-adic Numbers, p-adic Analysis, and Zeta-Functions (Second Edition). New York: Springer-Verlag, 1984.

Lang, Serge. Algebraic Number Theory. Massachusetts: Addison-Wesley, 1970.

Lorch, Edgar Raymond. Spectral Theory. New York: Oxford University Press, 1962.

Pedersen, Gert K. Analysis Now. New York: Springer-Verlag, 1989.

Pontryagin, L. Topological Groups. (Translated from the Russian by Emma Lehmer.) Princeton: Princeton University Press, 1939. Rudin, Walter. Real and Complex Analysis. New York: McGraw-Hill, 1966.

Ribes, L. Introduction to Profinite Groups and Galois Cohomology. Kingston, Ontario: Queen's University, 1970.

Rudin, Walter. Fourier Analysis on Groups. New York: John Wiley & Sons, 1962; Wiley Classics Library Edition, 1990.

Serre, J.-P. Corps locaux. Paris: Hermann, 1968.

Serre, J.-P. Abelian. l Adic Representations and Elliptic Curves (Second Edition). Massachusetts: Addison-Wesley, 1989.

Serre, J.-P. Galois Cohomology. (Translated from the French by Patrick Ion, with new additions.) New York: Springer-Verlag, 1997.

Shatz, S. Profinite Groups, Arithmetic and Geometry. Princeton: Princeton University Press, 1972.

Tate, J. FourierAnalysis in Number Fields and Hecke's Zeta Function. Thesis, Princeton University, 1950. Tate J. "Local Constants" in Algebraic Number Fields, ed. by A. Frohlich, New York: Academic Press, 1977.

Weil, Andrd. L'integration dans les grouper topologiques et ses applications. Paris: Herman, 1965.

Weil, Andr6. Basic Number Theory (Third Edition). New York: SpringerVerlag. 1974.

Suggestions for Further Reading To aid the reader we list below selected topics and corresponding references that are natural to pursue after this book. The list is not comprehensive, and we have certainly omitted some valuable and beautiful sources, especially where a heftier background is required.

References

341

-Topic 1: Lie Groups These ubiquitous topological groups are characterized by being locally Euclidean. Suggested texts are:

Chevalley, C. Theory Lie Groups, I. Princeton: Princeton University Press, 1946.

Dieudonne, J. Sur les groupes classiques. Paris: Hermann, 1973.

Serre, J.-P. Lie Groups and Lie Algebras (Second Edition). New York: Springer-Verlag, 1992. (This book also develops the parallel theory of padic analytic groups.)

-Topic 2: Topological Transformations Groups Koszul, J.-L. Lectures on Groups of Transformations. Bombay: Tata Institute of Fundamental Research, 1965. Montgomery, D and L. Zippin. Topological Transformation Groups. Huntington, New York: Robert Krieger Publishing Company, 1974. (This contains a discussion of Hilbert's fifth problem.) -Topic 3: Cohomology of Profinite Groups Serre, J.-P. Galois Cohomology. (See references above.)

-Topic 4: Unitary Representations Bruhat, F. Lectures on Lie Groups and Representations of Locally Compact Groups. (See references above.)

Knapp, A.W. Representation Theory of Semisimple Groups: An Overview Bases on Examples. Princeton: Princeton University Press, 1986.

-Topic 5: Discrete Subgroups of Lie Groups This subject is a vast generalization of the analysis of unit groups as lattices in Euclidean spaces. Borel, A. Introduction aux groupes arithmetiques. Paris: Hermann, 1969. Raghunathan, M.S. Discrete Subgroups of Lie Groups. New York: SpringerVerlag, 1972.

Zimmer, R. Ergodic Theory and Semisimple Groups. Boston: Birkhauser, 1984.

-Topic 6: Class Field Theory Artin, E. and J. Tate. Class Field Theory. New York: W. A. Benjamin, 1968. Lang, S. Algebraic Number Theory. (See references above.)

Langlands, R -P. "Abelian Algebraic Groups" in the Olga Taussky-Todd memorial volume of the Pacific Journal of Mathematics, 1998. (This work

342

References

treats the case of tori, the basic case to be understood before tackling the general philosophy of the author as it applies to the still open nonabelian case.) Serre, J.-P. Corps locaux. (See references above.) Weil, A. Basic Number Theory. (See references above.)

-Topic 7: Cyclotomic Fields and p-adic L-functions Iwasawa, K. Lectures on p-adic L -functions. Princeton: Princeton University Press, 1972. Lang, S. Cyclotomic Fields I and 11 (Combined Second Edition). New York: Springer-Verlag, 1990. de Shalit, E. Iwasawa Theory of Elliptic Curves with Complex Multiplication. New York: Academic Press, 1987. Washington, L. Cyclotomic Fields. New York: Springer-Verlag, 1982.

-Topic 8: Galois Representations and L-functions Serre, J.-P. Abelian ! Adic Representations and Elliptic Curves. (See references above.)

-Topic 9: The Analytic Theory of L-functions Apostol, T. Modular Functions and Dirichlet Series in Number Theory (Second Edition). New York: Springer-Verlag, 1990. Davenport, H. Multiplicative Number Theory. New York: Springer-Verlag, 1980.

Murty, Mr. R. and V. K. Murty. Nonvanishing of L functions and Applications. Boston: BirkhAuser, 1997. Siegel, C.L. On Advanced Analytic Number Theory (Second Edition). Bombay: Tata Institute of Fundamental Research, 1990. Titchmarsh, E. C. The Theory of the Riemann Zeta Function (Second Edition). Edited and with a preface by D.R. Heath-Brown. New York: Oxford University Press, 1986.

-Topic 10: SL2(R) and Classical Automorphic Forms Borel, A. Automorphic Forms on SL2(R). Cambridge: Cambridge University Press, 1997. Hida, H. Elementary Theory of L -functions and Eisenstein Series. Cambridge: Cambridge University Press, 1993. Iwaniec, H. Topics in Classical Automorphic Forms. Providence: American Mathematical Society, 1997. Lang, S. SL2(R). (Reprinted from 1975.) New York: Springer-Verlag, 1985.

References

343

Shimura, G. Introduction to the Arithmetic Theory of Automorphic Forms. (Reprinted from 1971.) Princeton: Princeton University Press, 1994.

Well, A. Dirlchlet Series and Automorphic Forms. New York: SpringerVerlag, 1971.

-Topic 11: Automorphic Forms via Representation Theory Bailey, T. N. and A. W. Knapp, eds. Representation Theory and A utomorphic Forms. Providence: American Mathematical Society, 1997. Borel, A. and W. Casselman, eds. Automorphic Forms, Representation and Lfunctions I, 11. Providence: American Mathematical Society, 1979. Bump, D. Automorphic Forms and Representations. Cambridge: Cambridge University Press, 1997. Gelbart, S. Automorphic Forms on Adele Groups. Princeton: Princeton University Press, 1975. Gelbart S. and F. Shahidi. Analytic Properties of Automorphic L -functions. Boston: Academic Press, 1988. Godement, R. Notes on Jacquet-Langlands Theory. (Mimeographed notes.) Princeton: Institute of Advanced Study, 1970. Jacquet, H. and R.P. Langlands. Automorphic Forms on GL(2). New York: Springer-Verlag, 1970. Langlands, R. P. Euler Products. New Haven: Yale University Press, 1971. FINAL REMARK. There are also connections of Tate's thesis with string theory. For instance, see:

Vladimirov, V.S. "Freund-Witten Adelic Formulas for Veneziano and Virasoro-Shapiro Amplitudes." Russian Mathematical Surveys, 48(6), 3-38, 1993.

Index

A

abelianization (of a group), 220 absolute norm, 336 absolute value(s) Archimedean, 157 definition, 154 equivalent, 156 adelic, 198 non-Archimedean, 157 normalized, 196 trivial, 156 ultrametric, 157 adele group, 189 adelic circle, 203 adjoint (of an operator), 62 admissible (adelic function), 260 Alaoglu's theorem, 322 almost everywhere, 323 approximation theorem, 190 Archimedean absolute value. See absolute value, Archimedean Artin's product formula, 198 Artin map, 224 functoriality, 225 Artin reciprocity law, 224 Artin symbol, 216 B

Banach algebra character, 56 definition, 50

quotient algebra, 55 Banach space, 47 Bochner's theorem, 111 Borel measure, 9 Borel subsets, 9 bounded away from zero, 68 bounded operator, 50 box topology, 21 C

63

canonical divisor (of a function field), 267 characters. See Banach algebra, local field, or topological group class number formula, 287, 288 classification theorem (for local fields), 140 commutator subgroup, 220 compact-open topology, 87 complete field, 157 complex measure, 71 conductor, 238, 253, 254 congruent to one (modulo an integral ideal), 206 connected component, 25 topological space, 25 convolution, 94 cyclotomic character, 131 cyclotomic polynomial, 228

346

Index

D

decomposition group, 165, 214 associated canonical homomorphism, 165 Dedekind domain, 165, 327 Dedekind zeta function, 278 degree (of a divisor), 265 degree-one prime, 216, 306 different, 176, 177, 254, 337 Dirac measure, 95 directed set, 19 Dirichlet's theorem (on primes in arithmetic progressions), 220, 293 Dirichlet character, 238 Dirichlet density, 293 lower and upper, 306 Dirichlet series, 242 with nonnegative coefficients, 290 discrete valuation (associated with a prime), 204 discrete valuation ring, 145 discriminant ideal, 176, 177, 337 of a basis, 337 division of places, 160 divisor (on a function field), 265 divisor class, 265 of degree zero, 266 divisor map, 265, 285 dual (with respect to the trace map), 254 dual measure, 103 dual subset (in a separable extension), 337 duality (for function spaces), 324 E

Eisenstein equation, 175 Eisenstein polynomial, 230 elementary functions, 98 epsilon factor, 246 essential supremum, 323 essentially bounded functions, 324 Euler's dilogarithm function, 305

Euler product expansion, 241 evaluation map, 319 exponent (of a character on a local field), 244

F Fermat equation, 213 finite total mass, 111, 127 first spectral theorem, 66 Fourier inversion formula, 103 finite version, 129 Fourier transform, 102 adelic, 260 finite version, 129 of a measure, 11 I of a Schwartz-Bruhat function, 246 fractional ideal, 204, 334 principal, 204, 334 Frobenius automorphism, 154 Frobenius class, 216 Frobenius element, 215 Frobenius map, 215 function field, 154 functional equation for the global zeta function, 271 for the local zeta function, 246 fundamental theorem of Galois theory, 34 G

Galois extension, 33 Galois group definition, 33 profinite topology, 34 gamma function, 244 Gauss sum, 243, 255 Gelfand topology, 58 Gelfand transform, 59, 64 Gelfand-Mazur theorem, 55 Gelfand-Naimark theorem, 63 genus (of a function field), 267 G-isomorphism, 50 G-linear map, 50 global field, 154

Index

Grossencharakter, 304

H Haar covering number, 12 Haar measure, 10 existence, 15 uniqueness, 16 Hecke's theorem, 303 Hecke character. See Grossencharakter Hecke L-function, 278 Heisenberg group, 38, 45, 211 Hensel's lemma, 170, 175 Hermitian operator. See self-adjoint operator Hilbert class field, 214 Hilbert space, 62 homogeneity (of a topological space), 2

ideal class group, 205, 334 ideal group, 334 idele class group, 196 idele group, 189 ideles of norm one, 200 induced measure (on a restricted direct product), 185 inertia group, 239 inner regular (measure), 10 integers (of a global field), 164 integrable functions, 323 integral closure, 329 integral elements, 328 inverse different, 176, 177, 337 inverse limit, 20 inverse system. See projective system involution, 63

K Krasner's lemma, 175 Kronecker's Jugendtraum, 227 Kronecker-Weber theorem, 227

347

L L-function, 242, 277 linear system (of a divisor), 266 local field, 133 characters, 243 local L-factor, 244 local ring, 145, 326 localization (of a ring), 326 locally compact group, 8 regular representation, 82 locally constant, 245 locally convex (topological space), 47 logarithmic map, 281

M maximal abelian extension, 226 maximal unramified extension, 219 measurable space, 9 module of a local field, 146 of an automorphism, 132 multiplicative subset, 326 N

natural density (of a set of primes), 220 neighborhood, 3

symmetric, 3 non-Archimedean absolute value. See absolute value, non-Archimedean norm homomorphism (on idele class groups), 223 norm map (on a field extension), 197, 336

norm of a bounded operator, 50 norm of a prime, 220 norm topology, 317 normal (operator), 62 normal extension, 33 normally convergent function, 260 nonmed linear spaces, 315 norm-one idele class group, 200 number field, 154

348

Index

0 order (of an element of a local field), 146

orthogonality of characters (for compact groups), 82 orthogonality relations (for compact groups), 81 Ostrowski's theorem, 158 outer regular (measure), 9

P p-adic integers, 24 Parseval's identity, 123 Pell's equation, 209 Peter-Weyl theorem, 84 Picard group, 196, 265 of degree zero, 266, 285 place(s) of a field, 156 finite, 164 infinite, 164 Plancherel's theorem, 122 Plancherel transform, 122 p-norm (on OJ, 144 Poisson summation formula, 262 polylogarithm function, 305 Pontryagin dual, 87 Pontryagin duality, 119 positive definite function. See positive type positive definite Hermitian form, 61 positive measure, 9 positive operator, 70 positive type (function of), 92 pre-Hilbert space, 61

preordered set, 19 pre-unitary (endomorphism or isomorphism), 73 prime (of a global field), 165 prime global field, 158 prime number theorem (for a number field), 312 principal divisor, 265 probability measure, 81 profudte group(s)

definition, 23 index (of a subgroup), 36 order, 37 structure, 31 topological characterization, 25 profrnite topology, 23 projective limit, 20 universal property, 20 projective system, 19

prop-group, 38 pro-p-Sylow subgroup, 39

Q quadratic residue, 213 quasi-characters, 243

R Radon integral, 323 Radon measure, 10, 323 ramification index, 152, 163, 335 ramified prime, 335 ray class group narrow, 207 wide, 206 regulator (of a number field), 283 regulator map, 282 representation(s) (topological), 47 (topologically) irreducible, 49 abstract, 47 algebraically irreducible, 49 equivalent, 50 induced, 84 multiplicity-free, 85 pre-unitarily equivalent, 74 pre-unitary, 73 unitarily equivalent, 74 unitary, 74 residual degree, 152, 163, 335 resolvent set, 52 restricted direct product characters, 182 definition, 180 restriction map (for places), 160

Index

Riemann hypothesis (for a function field), 312 Riemann zeta function, 241, 278 Riemann-Roch theorem, 264 geometric form, 267 root number, 242, 259 S

Schur's lemma, 75 Schwartz function, 245 Schwartz-Bruhat function, 246 adelic, 260 S-class group, 203

second spectral theorem, 72 self-adjoint function space, 60 self-adjoint operator, 62 self-dual measure, 245, 246, 300 separable (elements and extensions), 33

sesquilinear form, 72 shifted dual, 245 S-ideles, 201, 281 of norm one, 202 sigma compact (topological space), 324

sign character, 244 signed measure, 71 S-integers (of a global field), 202 smooth function (on a local field), 245 spectral measure, 71 spectral radius, 51 spectrum (of an element in a Banach algebra), 51 purely continuous, 85 standard character(s) adelic, 269 complex, 251 local non-Archimedean, 253, 297, 299 real, 249 Stone-Weierstzass theorem, 60 strictly multiplicative function, 137 supernatural number, 36

349

T tamely ramified extension, 177 Tauberian theorem (for Dirichlet series), 313 Tchebotarev density theorem, 220 theta function, 241 topological field, 46 topological group characters, 87 definition, I quotient space, 6 separation axioms, 5 topological vector space, 46 totally disconnected (topological space), 25 totally ramified extension, 152, 163 trace map (on a field extension), 336 transfer map, 221 on Galois groups, 223 transitivity, 222 transform topology, 107 translation (of functions), 4 translation-invariant Borel measure, 10 topology, 2 triangle inequality, 155

U

ultrametric absolute value. See absolute value, ultrametric ultrametric field or module, 137 ultrametric inequality, 137 uniform boundedness principle, 319 uniform continuity (left and right), 4 uniformizing parameter, 145, 327 unit ball, 315 unitary characters, 243 unitary operator, 62 unramified character, 244 unramified extension, 152, 163 unramified prime, 335

350

Index

V

Verlagerung. See transfer map

w weak dual (of a nonmed linear space), 318

weak topology, 317 weak-star topology, 58, 319

z zeta function global, 271 local, 246

Graduate Texts in Mathematics (conrowrd from page ii)

62

KARGAPOLOV/MERLZ3AKOV. Fundamentals

93 DuBRovmN/FoMENKo/Novtxov. Modem

Geometry-Methods and Applications.

of the Theory of Groups. 63 Bot.uwBAs. Graph Theory. 64 EDWARDS. Fourier Series. Vol. 12nd ed. 65 WEt.L.s. Differential Analysis on Complex Manifolds. 2nd ed. 66 WATERHOUSE. Introduction to Affine

Part 1. 2nd ed.

94 WARNER. Foundations of Differentiable Manifolds and Lie Groups. 95 96

SHIRYAEV. Probability. 2nd ed.

CONWAY. A Course in Functional

71

FARKAS/KRA. Riemann Surfaces. 2nd ed.

Analysis. 2nd ed. KoaLtrz. Introduction to Elliptic Curves and Modular Forms. 2nd ed. 98 BROcKER/Tom DtECK. Representations of Compact Lie Groups. 99 GROVE/BENSON. Finite Reflection Groups. 2nd ed.

72

STIILWE1L. Classical Topology and

100 BERG/Ctuus ENSEN/RESSEL. Harmonic

Group Schemes.

67 SERRE. Local Fields. 68 WEIDMANN. Linear Operators in Hilbert Spaces. 69 LANG. Cyclotomic Fields Ii. 70 MASSEY. Singular Homology Theory.

97

Analysis on Semigroups: Theory of

Combinatorial Group Theory. 2nd ed. 73 HUNG ruroi D. Algebra. 74 75

Positive Definite and Related Functions.

DAVENPORT. Multiplicative Number Theory. 2nd ed. HOCHSCHILD. Basic Theory of Algebraic

101 EDWARDS. Galois Theory. 102 VARADARAJAN. Lie Groups, Lie Algebras and Their Representations.

Groups and Lie Algebras.

103 LANG. Complex Analysis. 3rd ed.

76 77

IITAKA. Algebraic Geometry. HEcKE. Lectures on the Theory of Algebraic Numbers.

104 DUBROVIN/FOMENKO/NOVIKOV. Modem

78

BURRIS/SANKAPPANAVAR. A Course in

105

79

Universal Algebra. WALTERS. An Introduction to Ergodic Theory.

80 RoBINSON. A Course in the Theory of Groups. 2nd ed. 81

FORSTER. Lectures on Riemann Surfaces.

82

BOTr/Tu. Differential Forms in Algebraic Topology.

83 WASHINGTON. Introduction to Cyclotomic

Fields. 2nd ed. 84

IRELAND/ROSEN. A Classical Introduction

to Modem Number Theory. 2nd ed. 85 EDWARDS. Fourier Series. Vol. IT. 2nd ed.

86 VAN LINT. Introduction to Coding Theory. 2nd ed.

87 BROWN. Cohomology of Groups. 88 89

PIERCE. Associative Algebras.

LANG. Introduction to Algebraic and Abelian Functions. 2nd ed. 90 BRNNDSTED. An Introduction to Convex 91

Polytopes. BEARDON. On the Geometry of Discrete

Groups. 92 DIESTEL. Sequences and Series in Banach Spaces.

Geometry-Methods and Applications. Part 11.

LANG. Slr(R). 106 Stt.vaRMAN. The Arithmetic of Elliptic Curves.

107 OLVER. Applications of Lie Groups to Differential Equations. 2nd ed. 108 RANGE. Holomorphic Functions and Integral Representations in Several

Complex Variables. 109 LEirro. Univalent Functions and Teichmiiller Spaces. 110 LANG. Algebraic Number Theory. I I I HUSEMOLIER. Elliptic Curves.

112 LANG. Elliptic Functions. 113 KARArzAS/SHREVE. Brownian Motion and Stochastic Calculus. 2nd ed. 114 KOBLI Z. A Course in Number Theory and Cryptography. 2nd ed.

115 BERGER/GosTu ux. Differential Geometry: Manifolds, Curves, and Surfaces. 116 KEU.EY/SREJtvASAN. Measure and

Integral. Vol. 1. 117 SERRE. Algebraic Groups and Class Fields.

118 PEDERSEN. Analysis Now.

119 ROTMAN. An Introduction to Algebraic Topology.

120

ZIEMER. Weakly Differentiable Functions:

Sobolev Spaces and Functions of Bounded Variation. 121 122

LANG. Cyclotomic Fields I and H. Combined 2nd ed. REMMERT. Theory of Complex Functions. Readings in Mathematics

123 EBBINGHAUS/HERMES et al. Numbers.

Readings in Mathematics

149 RATCLIFFE. Foundations of Hyperbolic Manifolds. 150 EISENBUD. Commutative Algebra

with a View Toward Algebraic Geometry. 151

SILVERMAN. Advanced Topics in

the Arithmetic of Elliptic Curves. 152 ZIEGLER. Lectures on Polytopes.

153 FULTON. Algebraic Topology: A

124 DUBROVIN/FOMENKO/NOVIKOV. Modem

First Course. BROWN/PEARCY. An Introduction to

Geometry-Methods and Applications.

154

Part 111.

Analysis. 155 KASSEL. Quantum Groups.

125 BERENSTEIN/GAY. Complex Variables: An Introduction.

126 BOREL. Linear Algebraic Groups. 2nd ed. 127 MASSEY. A Basic Course in Algebraic Topology. 128 RAUCH. Partial Differential Equations.

156 KECHRIS. Classical Descriptive Set Theory.

130 DODSON/POSTON. Tensor Geometry.

157 MALLIAVIN. Integration and Probability. 158 ROMAN. Field Theory. 159 CONWAY. Functions of One Complex Variable II. 160 LANG. Differential and Riemannian Manifolds.

131 LAM. A First Course in Noncommutative

161

Rings. 132 BEARDON. Iteration of Rational Functions. 133 HARRIS. Algebraic Geometry: A First Course. 134 ROMAN. Coding and Information Theory. 135 ROMAN. Advanced Linear Algebra.

Polynomial Inequalities. 162 ALPERIN/BELL. Groups and Representations. 163 DIXON/MORTIMER. Permutation Groups. 164 NATHANSON. Additive Number Theory:

136 ADKINSIWEINTRAUB. Algebra: An

The Classical Bases. 165 NATHANSON. Additive Number Theory: Inverse Problems and the Geometry of Sumsets. 166 SHARPE. Differential Geometry: Cartan's

129 FuLTON/HARRIS. Representation Theory:

A First Course. Readings in Mathematics

Approach via Module Theory. 137 AXLER/BOURDONIRAMEY. Harmonic

Function Theory. 138 COHEN. A Course in Computational Algebraic Number Theory. 139 BREDON. Topology and Geometry. 140 AUBIN. Optima and Equilibria. An Introduction to Nonlinear Analysis. 141 BECKER/WEISPrcNNLNG/KREDEL. Grobner

Bases. A Computational Approach to Commutative Algebra. 142 LANG. Real and Functional Analysis. 3rd ed.

143 DOOB. Measure Theory. 144 DENNIS/FARB. Noncommutative Algebra.

145 VICK. Homology Theory. An Introduction to Algebraic Topology. 2nd ed. 146 BRIDGES. Computability: A Mathematical Sketchbook. 147 ROSENBERG. Algebraic K-Theory and Its Applications. 148 ROTMAN. An Introduction to the Theory of Groups. 4th ed.

BORWEIN/ERD);LYI. Polynomials and

Generalization of Klein's Erlangen Program. 167 MORANDI. Field and Galois Theory. 168 EWALD. Combinatorial Convexity and

Algebraic Geometry. 169 BHATIA. Matrix Analysis. 170 BREDON. Sheaf Theory. 2nd ed. 171 PETERSEN. Riemannian Geometry. 172 REMMERT. Classical Topics in Complex Function Theory. 173 DIESTEL. Graph Theory.

174 BRIDGES. Foundations of Real and

Abstract Analysis. LICKORISH. An Introduction to Knot Theory. 176 LEE. Riemannian Manifolds. 177 NEWMAN. Analytic Number Theory. 175

178 CLARKE/LEDYAEV/STERN/WOLENSKI.

Nonsmooth Analysis and Control Theory.

179 DOUGLAS. Banach Algebra Techniques in Operator Theory. 2nd ed.

180 SRIVASTAVA. A Course on Borel Sets.

186 RAMAKRISHNANNALENZA. Fourier

KRESS. Numerical Analysis. 182 WALTER. Ordinary Differential Equations. 183 MEGGINSON. An Introduction to Banach

187 HARRIS/MORRISON. Moduli of Curves. 188 GOLDBLATr. Lectures on the Hyperreals:

181

Space Theory.

184 BOLLOBAS. Modern Graph Theory. 185 Cox/LITrLEIO'SHEA. Using Algebraic Geometry.

Analysis on Number Fields.

An Introduction to Nonstandard Analysis. 189 LAM. Lectures on Modules and Rings.

The general aim of this book is to provide a modem approach to number theory through a blending of complementary algebraic and analytic perspectives, emphasizing harmonic analysis on topological groups. The more particular goal is to cover John Tate's visionary thesis, giving virtually all of the necessary analytic details and topological preliminaries-technical prerequisites that are often foreign to the typical, more algebraically inclined number theorist. While most of the existing treatments of Tate's thesis are

somewhat terse and less than complete, the authors' intent is to be more leisurely, more comprehensive, and more comprehensible. The text addresses students who have taken a year of graduate-level courses in algebra, analysis, and topology. While the choice of objects and methods is naturally guided by specific mathematical goals, the approach is by no means narrow. In fact, the subject matter at hand is germane not only to budding number theorists, but also to students of harmonic analysis or the representation theory of Lie groups. Moreover, the work should be a good reference for working mathematicians interested in any of these fields. Specific topics include: topological groups, representation theory, duality for locally compact abelian groups, the structure of arithmetic fields, ade-

les and ideles, an introduction to class field theory, and Tate's thesis and applications.

ISBN 0-387-48436-4

ISBN 0-387-98436-4 www.springer-ny.com

.7i;103l7

'9plI3III,

Fourier Analysis on Number Fields (Graduate Texts in Mathematics) (v. 186)

Fourier analysis on number fields

Local Fields (Graduate Texts in Mathematics 67)

Cyclotomic Fields (Graduate Texts in Mathematics)

Cyclotomic Fields II (Graduate Texts in Mathematics)

Local Fields (Graduate Texts in Mathematics)

Local Fields (Graduate Texts in Mathematics)

Algebraic Number Theory (Graduate Texts in Mathematics)

Modern Fourier Analysis, Second edition (Graduate Texts in Mathematics)

Analytic Number Theory (Graduate Texts in Mathematics)

Analytic Number Theory (Graduate Texts in Mathematics)

Classical Fourier Analysis, Second Edition (Graduate Texts in Mathematics, 249)

Modern Fourier Analysis, Second edition (Graduate Texts in Mathematics)

Classical Fourier Analysis, Second Edition (Graduate Texts in Mathematics, 249)

Algebra (Graduate Texts in Mathematics) (v. 73)

Fourier Analysis (Graduate Studies in Mathematics)

Algebra (Graduate Texts in Mathematics) (v. 73)

Fourier Analysis (Graduate Studies in Mathematics)

Elements of Functional Analysis (Graduate Texts in Mathematics) (v. 192)

Lectures on discrete geometry (Graduate Texts in Mathematics v.212)

Analysis Now (Graduate Texts in Mathematics)

Matrix Analysis (Graduate Texts in Mathematics)

Matrix Analysis (Graduate Texts in Mathematics)

Analysis Now (Graduate Texts in Mathematics)

Lectures on Discrete Geometry (Graduate Texts in Mathematics v.212)

Analysis Now (Graduate Texts in Mathematics)

Elementary Functional Analysis (Graduate Texts in Mathematics)

Analysis Now (Graduate Texts in Mathematics)

Matrix Analysis (Graduate Texts in Mathematics)

Analysis Now (Graduate Texts in Mathematics)

Analysis Now (Graduate Texts in Mathematics)

Fourier Analysis on Number Fields (Graduate Texts in Mathematics) (v. 186)